JP3130934B2 - Variable bit rate audio encoder - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a U.S. application Ser.
This is a continuation-in-part of US application SN 07 / 665,948, filed March 6, 1991, which is a continuation application of 07 / 664,579.

BACKGROUND OF THE INVENTION The present invention relates to the transmission of audio and video information over a channel of fixed capacity, such as a telephone communication channel.

Video conferencing systems typically transmit both audio and video information on the same channel. A portion of the channel bandwidth is usually dedicated to audio information, and the remaining bandwidth is allocated to video signals.

The amount of video and audio information changes with time. As an example, a person at one end of the system may be silent at some point. Thus, if the system includes a variable capacity voice encoder, during this period of silence, very little information needs to be transmitted.

Similarly, there may be little or no change in video information between frames, such as when all objects in the field of view are stationary. If the system has a variable capacity video encoder, there is little information to be transmitted during these inactive periods. At the other extreme, during periods of high activity, the amount of video information may exceed the channel capacity allocated to the video information. Therefore, the system transmits as much video information as possible and discards the rest.

Typically, video encoders give priority to the most salient features of a video signal. Thus, high-priority information is transmitted first and unobtrusive low-priority information is temporarily discarded if the channel does not have sufficient capacity. It is therefore desirable to have as much bandwidth available as possible.

Accordingly, it is an object of the present invention to reduce the amount of channel bandwidth allocated to audio information when there is little audio information that needs to be transmitted.
The rest of the bandwidth is allocated to video information. Thus, on average, the bit rate for audio information is low and the bit rate for video information is high.

SUMMARY OF THE INVENTION The present invention relates to a method and apparatus for allocating transmission bits used to transmit samples of a digital signal. For a frame consisting of a plurality of samples of the digital signal, a total permissible quantization distortion value representing an aggregate permissible quantization distortion error is selected. For a frame of multiple samples, a set of samples is selected such that the selected frames are greater than a noise threshold. For each sample in the set, a sample quantization distortion value representing an allowable quantization distortion error for that sample is calculated. The sum of all sample quantization distortion values is approximately equal to the collective allowable quantization distortion value. For each sample in the set, select a quantization step size that gives a quantization distortion error approximately equal to the corresponding quantization distortion value of the sample. Next, each sample is quantized using the quantization step size.

According to a preferred embodiment, the digital signal includes a noise component and a signal component. A signal index representing the ratio of the value of the signal component to the value of the noise component is created for at least one sample of the frame. Based on the signal index, a collective allowable quantization distortion is selected.

The sample quantization distortion value is calculated by dividing the collective allowable quantization distortion value by the number of samples in the frame to form a first sample distortion value. A trial set of samples of the digital signal is selected such that each sample of the set is greater than a noise threshold determined at least in part by the value of the first sample distortion value. The first sample distortion value is adjusted by an amount determined by the difference between the first sample distortion value and at least one sample removed from the trial set (ie, a "noisy sample"). A trial set of noisy (noisy) samples is identified based on the adjusted distortion values, the noisy samples (if any) are removed from the trial set, and a first sample distortion value and a noisy sample are determined. The steps of re-adjusting the first sample distortion value by an amount determined by the difference between are repeated. An adjusted first such that no additional noisy samples in the trial set are found.
The process is repeated until the sample distortion value is reached or the process is repeated a maximum number of times.

After the adjustment is completed, the number of bits required to transmit all the samples in the trial set is estimated. The estimated number of bits is compared to the maximum number of bits. If the estimated number of bits is less than or equal to the maximum number of bits, a final noise threshold is selected based on the adjusted first sample distortion value.

If the estimated number of bits exceeds the maximum number of bits, a second sample distortion value is created. A second trial set of all samples of the digital signal is selected such that each of the plurality of samples of the set has a value greater than the second sample distortion value. The number of bits needed to transmit the sample is then estimated. The estimated number of bits is then compared to the maximum number of bits. If the estimated number of bits is greater than the maximum number of bits, the second sample distortion value is increased, and a second trial set of samples is re-created based on the adjusted second sample distortion value. select. Re-estimate the number of bits needed to transmit the second trial set. This process is repeated until the estimated number of bits reaches a second sample distortion value that is less than or equal to the maximum number of bits.

Next, a sample distortion value is calculated from the adjusted first and second sample distortion values. A final set of samples of the digital signal is selected such that each of the final set of samples has a value greater than a final threshold determined by the corresponding sample distortion value of the sample.

According to another aspect, the present invention relates to a method and apparatus for communicating a digital signal including a noise component and a signal component.
An estimated signal having fewer samples than the digital signal represented by the digital signal is created. A signal index is created that represents the ratio of the value of the signal component to the value of the noise component for at least one sample of the estimated signal. A plurality of samples of the digital signal having a sufficiently large signal component are created based on the signal index. The selected plurality of samples of the digital signal and the samples of the digital estimated signal are all transmitted to a remote device. The remote unit reconstructs a digital signal from the selected and transmitted plurality of samples and the estimated samples.

According to a preferred embodiment, the digital signal is a frequency domain audio signal representing the audio information to be communicated, and each sample of the estimated signal is a spectral estimate of the frequency domain signal in the corresponding frequency band. To reconstruct the frequency domain signal, a random number is generated for each unselected sample of the frequency domain audio signal. A noise estimate of the magnitude of the noise component of the unselected sample is created from at least one spectral estimate. Generate a scaling factor based on the noise estimate. The random number is then scaled according to a scaling factor to create a reconstructed sample representing the unselected sample.

The estimated signal and the frequency domain audio signal each comprise a series of frames. Each frame represents audio information over a specified time window. To generate an initial noise estimate for the current frame, an initial noise estimate is first generated for a previous frame with an estimated signal. Based on the value of the signal index, selecting a rise time constant t _r and time fall constant t _f. The selected rise time constant is added to the first noise estimate to form an upper threshold, and the selected fall time constant is subtracted from the first noise estimate to form a lower threshold.

Next, the current spectral estimate of the current frame representing the same frequency band is compared to upper and lower thresholds. If the current spectral estimate is in the middle of the threshold,
Let the current noise estimate be equal to the current spectral estimate. If the current spectrum estimate is greater than the upper threshold, the current noise estimate is set equal to the upper threshold.

To generate a scaling factor, a noise factor and a speech index are first selected based on the value of the signal index.
Next, the current noise estimate is multiplied by the noise figure. Similarly, multiply the current spectral estimate by the signal index.
The results of these multiplications are added together to form a scaling factor from the sum.

According to another aspect, the present invention relates to a method and apparatus for estimating the energy of a speech component of each of a series of frames of a digital signal. A first frame noise estimate is generated that represents noise energy in a first frequency band of the first frame. A signal energy value representing the energy of the digital signal in the first frequency band of the second frame is also formed. If the signal energy value is greater than the first frame noise estimate, a small increment is added to the first frame noise estimate to form a second frame noise estimate. If the energy value is less than the first frame noise estimate, a large decrement is subtracted from the second frame noise estimate to form a second frame noise estimate. The second frame noise estimate is subtracted from the signal energy in the first frequency band to form a speech estimate representing the energy of the speech component in the first frequency band of the second frame.

Other objects, features and advantages of the present invention will become more apparent from the preferred embodiments described below with reference to the drawings.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1A is a block diagram of a near end of a video conference system.

FIG. 1B is a block diagram of the far end of the video conference system.

FIGS. 2A to 2C are diagrams showing estimated values of one set of frequency coefficients and two kinds of frequency coefficients.

3 (a) and 3 (b) are flowcharts illustrating a process for calculating a spectrum estimation value.

FIG. 4 is a diagram illustrating a set of spectral estimates configured for broadband.

 FIG. 5 is a block diagram showing a voice detection device.

FIG. 6 is a flowchart showing a method for estimating the amount of sound in a frame of a microphone signal.

 FIG. 7 is a block diagram of the bit rate estimator.

FIGS. 8A and 8B are flowcharts showing a method for calculating a first estimated value of allowable distortion per spectrum estimated value.

FIG. 9 is a diagram showing a method for calculating a second estimated value of the allowable sum per spectrum estimated value.

FIG. 10 is a flowchart showing a method of calculating the quantization step size.

FIG. 11 is a diagram illustrating interpolating between band quantization step sizes to form a coefficient quantization step size.

FIG. 12 is a block diagram showing a coefficient filling module.

DESCRIPTION OF THE PREFERRED EMBODIMENTS System Overview In FIGS. 1 (a) and 1 (b), a video conferencing system receives the voice of a person speaking at the near end and an electronic microphone signal m (t) representing that voice. (Where t is a variable representing time). Similarly, the camera 14 is focused on the talking person to generate a video signal v (t) representing an image of the person.

The microphone signal is supplied to an audio encoder 16 that digitizes and encodes the microphone signal m (t). As described in more detail below, encoder 16 includes a set of digitally encoded signals F _h (k), S _h (j), g _i (F), and A _i (F), which together represent the microphone signal.
Generate. As described in detail below, k and j are integers representing frequency values, and F is an integer specifying a “frame” of the microphone signal. Similarly, the video signal v
(T) is supplied to a video encoder 18 for digitizing and encoding the video signal.

The encoded video signal V _e and the encoded set of microphone signals are provided to a bit stream controller 20 that merges the signals into a serial bit stream for transmission on a transmission channel 22. .

Referring to FIG. 1B, the far-end bit stream controller 24 receives the bit stream from the transmission channel 22 and separates it into a video component and an audio component. The received video signal _Ve is converted to a video decoder
It is decoded by 26 and supplied to the display device 27 to reproduce the display of the near-end camera image. At the same time,
An audio decoder 28 decodes the received set of microphone signals and provides a near-end microphone signal m (t).
Is generated. The speaker 29 reproduces near-end sound in response to the speaker signal.

In FIG. 1 (a), encoder 16 includes an input signal conditioner 30 that digitizes and filters microphone signal m (t) to produce digitized microphone signal m (n). Here, n is an integer indicating the time point. The digitized microphone signal m (n) separates m (n) into a plurality of groups m _p (n) of 512 consecutive samples (referred to as frames) in the example shown. Each frame is selected to overlap each other. More specifically, each frame contains the last 16
It has a sample and 496 new samples.

Each sample frame is provided to a normalization module 34 which calculates the average energy _Eav of all microphones in the frame.

Module 34 then calculates, for each frame F, a normalized gain g (F) equal to the square root of the average energy E _av of that frame.

The normalized gain g is used at the near end to scale the microphone samples during each frame, as described in more detail below. This normalization gain is used at the far end to return the decoded microphone samples to their original scale. Therefore, the same gain used to scale the samples at the near end must be transmitted to the far end for use for rescaling the microphone signal. Since the gain g calculated by module 34 has a relatively large number of bits, module 34 uses a smaller number of bits than the number of bits to specify gain g
Quantizer that generates a gain quantization index g _i representing
Supply gain g to 35. The gain quantization index g _i is
It is then provided to bitstream controller 20 for transmission to the far end.

The far end audio decoder 28 reconstructs the gain from the transmitted gain quantization index (gain quantization index) g _i , returning the microphone signal to its original scale from the constructed gain g _q . Since the reconstructed gain _gq is typically slightly different from the original gain g, the near-end audio encoder 16 uses the same gain _gq as used at the far end to Normalize the signal. More specifically, the inverse quantizer is a far-end audio decoder
Similarly to 28, the gain quantization index g _i to the gain g _q
Reconfigure.

Quantization gain g _q and microphone signal frame m _F (n)
Is transferred to a discrete cosine transform module (DCT) 36, which divides each microphone sample m _F (n) by a gain g _q to form a normalized sample m ′ (n). The DCT 36 then uses the well-known discrete cosine variable algorithm to normalize the microphone signal m '
The group of (n) is converted into a normalized microphone signal. DCT3
6 thus generates 512 frequency coefficients F (k) representing samples of the frequency spectrum of the frame.

The frequency coefficient F (k) is encoded (using an entropy adaptive transfer coder described below) and transmitted to the far end. Encoder 16 may reduce the number of bits required to transmit these coefficients by estimating the relative amounts of signals (e.g., speech and noise) in various regions of the frequency spectrum, and generating a relatively large amount of noise. In addition, the encoder 16 chooses not to transmit the coefficients for the frequency bands containing the coefficients, and the encoder 16 determines the number of bits required to represent the coefficients, depending on the amount of noise present in the frequency band containing the frequencies of the coefficients. Is selected for each coefficient selected for transmission.More specifically, as is well known, one can tolerate more interference noise in regions of the audio spectrum that have a relatively large amount of audio signal. It is known that the audio signal has a tendency to mask noise, so that the encoder
16 coarsely quantizes a relatively large amount of audio signal.
This is because the quantization distortion introduced by the coarse quantization masks the audio signal. Therefore, the encoder 16 minimizes the number of bits required to represent each coefficient by selecting a quantization step size that matches the amount of audio signal represented by the coefficient.

To estimate the relative amounts of noise and signal in various regions of the spectrum, the frequency coefficient F (k) is
First, it is supplied to the spectrum estimation module 38. As described in detail below, the frequency coefficient F (k) is reduced to a smaller number of spectral estimates representing the frequency spectrum of the frame with less detail (detail) (j is the frequency band in the spectrum). ). The voice detection module 40 processes the spectral estimates of each frame to estimate the energy component of the microphone signal due to noise in the near-end room. Module 40 then provides a signal index A _i (F) that approximates the percentage (%) of the microphone signal in the frame to be attributed to the audio information.

Signal index A _i (F) and spectral estimate S
Both (j) are used to quantize and encode the frequency coefficient F (k) transmitted to the far end. Therefore,
These are both needed at the far end to reconstruct the frequency coefficients. Signal index A _i (F)
Has only three bits and is therefore provided directly to the bit stream controller 20 for transmission. However, the spectral estimate S (j) is
g ₂ S ² (j), and in order to reduce the number of bits to be transmitted, a well-known differential pulse code modulation encoder (DPC)
M) 39. (The DPCM is a first order DPCM with a unity gain predictor coefficient). The encoded log spectrum estimate S _e (j) is further encoded using a Huffman encoder 49 to further reduce the number of bits to be transmitted. The resulting Huffman code S
_h (j) is provided to the bitstream controller 20 for transmission.

The far-end audio decoder 28 receives the Huffman code S _h (j)
Reconstruct the log spectrum estimate Log ₂ S ² (j) from
However, due to the operation of the DPCM encoder 39, the reconstructed spectral estimate Log ₂ S _q ² (j) becomes the original estimate Log ₂ S
² Not the same as (j). For this reason, the near-end decoder 41
Decodes the encoded estimate S _e (j) in the same way as that made at the far end, and converts the decoded estimate Log ₂ S _q ² (J) in this way to the frequency coefficient F ( k)
Is used to quantize and encode. Thus, as described in more detail below, audio encoder 16 outputs signal Log ₂ S _q
^{2 The} same estimate Lo used by far end audio decoder 28 to reconstruct (j) at the far end
The frequency coefficient is encoded using g ₂ S _q ² (j). The bit rate estimator 42 calculates the decoded spectrum estimate
Based on the value of Log ₂ S _q ² (j) and the signal index A _i (F), a group of frequency coefficients having a certain sufficient amount of audio information as a whole worth transmitting to the far end is selected. The bit rate estimator then selects a quantization step size for each coefficient to be transmitted, which determines the number of bits used to quantize the coefficient for transmission. A bit rate estimator for each selected group of frequency coefficients
42 first calculates the group quantization step size Q (j) and "class" C (j) (described below) for each frequency band j. Next, the quantization step size is a coefficient quantization step size Q for each frequency coefficient F (k).
It is then interpolated to give (k). Q (k) is
Next, it is supplied to a coefficient quantizer 44, and the coefficient quantizer 44
Based on the assigned step size, each frequency coefficient in the band is quantized to provide a corresponding quantization index I (k).

The limit controller 45 generates a Huffman index I _h (k) in response to each quantization index I (k). The Huffman index I _h (k) supplied by the limit controller is then encoded by a Huffman encoder 47 to give a Huffman code F _h (k). The Huffman code is provided to bitstream controller 20 for transmission over transmission channel 22.

The limit controller uses a low frequency quantization index (ie, k
= 0) and starts the encoding process until all exponents are encoded or the number of encoded bits exceeds the capacity of channel 22 allocated to the microphone signal. Continue the process. When limit controller 45 detects that the capacity has been exceeded, it discards (discards) the remaining exponents I _h (k) and displays a unique Huffman signal indicating that the remaining frequency coefficients of the frame are not encoded for transmission. Send out the index I _h (k). (As described below, the far-end decoder estimates the uncoded coefficients from the transmitted spectral estimate).

Spectral estimate S for calculating the spectral estimate
(J) Referring to FIGS. 2 (a-c) and 3 (ab), next,
The spectrum estimation module 38 will be described in detail. Module 38 first divides frequency coefficient F (k) into L adjacent coefficient bands (L is preferably 10).
(Step S110). Module 38 provides, for each band j, a first approximation sample representing the entire band by calculating the spectral energy in that band. More specifically, module 38 squares (squared) each frequency coefficient F (k) in the band and adds the parallel (squared) of all coefficients in the band (steps 112, 114).
(See also FIGS. 2 (b) and 2 (c)).

This approximation only gives a poor representation of the spectrum if the spectrum contains high energy tones at the boundary between adjacent bands. For example, the spectrum shown in FIG. 2B includes a tone 50 at the boundary between the bands 52 and 54. Approximate sample 58 (representing the sum of squares of all coefficients in band 54) and approximate sample 56 (square (2
Interpolation between (representing the sum of the powers) yields a value 59 that does not accurately reflect the presence of tone 50.

Therefore, module 38 also uses a second approximation approach where the spectral estimates for both bands 52-54 reflect the presence of tone 50 near the boundary between the bands.

More specifically, module 38 derives a second approximation for each band using 10 samples in the band and 5 adjacent samples in each adjacent band (FIG. 2 (a)). reference). The module 38 performs a logical inclusive OR operation on the binary values of all 20 samples (step 116). This operation gives a computationally inexpensive estimate of the size of the largest sample in the set. More specifically, each digit of the binary value resulting from the "OR" operation is set to 1 if the operand binary value has a digit of 1 (eg, 0110 "OR" 0011).
= 011). Thus, the result of the "OR" operation is at a minimum equal to the value of the largest sample in the set, and at most twice as large. Next, the result of the "OR" operation is doubled to give a second approximation (step 117).

As shown in FIG. 2A, the second approximation in band 52 includes tone 50 from band 54, thus giving a relatively large approximation 60 for band 52. The second approximation 60 reflects the presence of the tone 50 more accurately than the first approximation 56. Therefore,
Module 38, a second approximation as compared to the first approximation (step 118), square towards magnitude of the two parties (the square) is selected as the spectral estimate S ² (j) (step 120-12
2). Module 38 finally calculates the logarithm of the square (square) spectral estimate Log ₂ S ² (j) (step 22).
Four).

Voice Detector for Calculating Signal Index A _i Referring to FIG. 4, the voice detector 40 calculates a signal index A _i (F) indicating a relative amount of voice energy in a frame.
Detector 40 groups the samples of the spectral estimates into multiple frequency broadbands for this purpose. These broadbands have variable bandwidths that are unevenly distributed over the spectrum. As an example, the first broadband may be wider than the second broadband, and the second broadband may be wider than the third broadband.

Referring to FIG. 5, the speech detector, to estimate the amount of background noise S _n in each broadband. For each broadband, speech detector 40 forms a collective (aggregate) estimate S _a (F) as follows.

Where F is an integer identifying the current frame, X and Y
Are the upper and lower frequencies of the broadband (step 210). As described in detail below, the voice detector 40
The collective estimate of the current frame is compared with the collective estimate of the previous band from the previous frame to determine the amount of noise in the band (step 211). In general, frames with relatively low approximate estimates often contain little voice energy in broadband. For this reason,
The collective estimate of this frame gives a reasonable estimate of the background noise in broadband.

The speech detector must first normalize each collective estimate in order to compare the estimate to a noise threshold. Otherwise, the estimate from the current frame will scale differently than the estimate from the previous frame, making the comparison meaningless. For this reason, the sound detector
In order to keep the normalized gain on the same log scale as the collective estimate, the collective estimate is unnormalized by first calculating Log ₂ (g _q ² ) (step 21).
2).

The denormalized estimate is compared to an upper threshold S _r (F) and a lower threshold S _f (F), where F represents the current frame (step 214). As described in more detail below, these thresholds are calculated for each frame based on an estimate of the noise from previous frames. Since the first frame has no previous frame, the upper threshold S _r (0) of the first frame is initialized to a value “r”, and
S _f (0) is initialized to −f. Where f is substantially greater than r (eg, r = 1, f = 10).

If the collective estimate S _a (F) is between the thresholds, the speech detector sets the estimate of noise from the frame equal to the collective estimate (step 216).

S _n (F) = S _a (F) (4) If the collective estimate is greater than the upper threshold, the noise estimate is set equal to the upper threshold (step 21).
8).

S _n (F) = S _r (F) (5) Finally, if the collective estimate is smaller than the lower threshold,
The noise estimate is set equal to the following threshold (step 220).

S _n (F) = S _f (F) (6) Before calculating the noise estimate for the next frame, the speech detector determines the thresholds S _f (F + 1), S _r (F +
1) to adjust the noise estimate S _n (F) of the current frame
(Step 222). More specifically, for the next frame F = F + 1, the speech detector calculates from the noise estimate of the current frame at the upper noise threshold by the following equation: S _r (F + 1) = S _n (F) + r (7) .

Similarly, the voice detector calculates the lower threshold value S _f (F + 1) = S
Calculate _n (F) -f.

Thus, for each new frame, the speech detector calculates a noise estimate and adjusts the upper and lower noise thresholds to span the noise estimate.

This technique adjusts the noise estimate over time so that the noise estimate in a broadband of the current frame is the most recent of the broadband.
Adjust adaptively to get the minimum set estimate of t). For example, if a series of frames without audio components arrive in broadband, the collective estimate will be relatively small since it only reflects the presence of background noise. Thus, if these collective values are less than the lower threshold S _f, then the above approach will quickly reduce the noise estimate in relatively large increments f until the noise estimate is equal to the relatively low collective estimate. To reduce.

The audio detector can detect an increase in background noise by incrementing the noise level upward.
However, because a relatively small increment "r" is used, the noise estimate tends to stay near the most recent minimum collective estimate. This eliminates the possibility that the speech detector will mistake the transient speech energy for background noise.

After calculating the noise estimate for a frame, the speech detector subtracts the noise estimate S _n (F) from the collective estimate S _a (F) for that frame to obtain an estimate of the speech signal in broadband. (Step 224). The audio detector
The threshold constant S _T is subtracted from the collective estimate speech signal to generate a signal indicating the degree that exceeds the certain threshold S _T (step 224). Finally, the speech detector calculates the exponent A of all frames.
_i (F) is calculated from the S _out signal and the normalized gain g _q (step 226).

The details of the calculation of A _i (F) from the set of S _out signals will be described with reference to FIG. The voice detector selects the largest S _out from all broadbands (step 24).
6). If the selected value _Smax is less than or equal to 0, all broadband is likely to have no audio component (step 248). Thus, the speech detector sets the index A _i (F) to 0, indicating that the broadband contains only noise (step 250).

If S _max is greater than 0 (step 248), the speech detector
_An index A _i (F) is calculated from the value of _max . The audio detector
Scaling the fixed gain G _s and S _max for this purpose. That is, S ′ _max = S _max * G _s (G _s is preferably
0.2734) (step 252). Speech detector is then calculated display g _o attenuated corresponding normalization gain the following equation.

g _o = (Log ₂ g _q −T _g ) * G _g (8) Here, T _g and G _g are predetermined constants, for example, T _g = 4096,
There is G _g = 0.15625 (step 254). If the g _o is S _'max greater than, the speech detector, S _max is assumed to be smaller than the sound energy in the frame. Therefore speech detector selects g _o as index A _i (F) (step 256) otherwise selecting S _'max as index A _i (F) (step 256). The speech detector finally compares the selected index with the maximum index i _max (step 258). If the selected index exceeds the maximum index, the voice detector will
Let A _i (F) be equal to its maximum value i _max (step 26)
2). Otherwise, the selected value is used as an index (step 260).

Bit Rate Estimator for Calculating Step Size Q (k) and Classification Information C (j) Referring to FIG. 1 (a), the bit rate estimator 42 calculates the index A _i (F) and the quantized Log spectrum estimate Log ₂
Receiving S _q ² (j), and in response, calculating a step size Q (k) for each frequency coefficient and a class indicator c (j) for each band j of frequency coefficients.
The class indicator c (j) is used by Huffman encoder 47 to select an appropriate code table.

The procedure used by bit rate estimator 42 to calculate step size and class information is described in more detail below. Referring to FIG. 7, the bit estimator comprises a first table 70 containing a predetermined total allowable distortion value _DT for each value of the signal index A _i (F). Each value _DT represents the collective quantization distortion error for all frames. The second table 72 stores a predetermined maximum number of bits r _max initially assigned to the frame for each location of the index A _i (F) (as will be described later, more if necessary). Bits can be assigned to frames). The stored bit rate r _max increases with A _i (F). Conversely, the stored distortion value D _T is A _i (F)
Decreases with each increase in. As an example, if A _i (F) is equal to 0 (ie 100% noise), the table gives a low bit rate and a high tolerable distortion. A _i
If (F) is equal to 7 (ie 100% audio), the table gives a high bit rate and low tolerable distortion. The first table and the second table select the collective allowable quantization distortion D _T and the allowable maximum number of bits r _max based on the value of the index A _i (F). This allowable distortion _DT is supplied to the first distortion approximation module 74. Module 74 then:
Calculate a _first allowable sample distortion value d1 representing the allowable distortion per estimate. Then using a sample distortion value d ₁ c
(J) and the block quantization step size Q (k) are derived.

Calculation of First Approximation of Allowable Distortion d _{1 With} reference to FIGS. 8A and 8B, the module 74 includes:
By dividing the allowable total distortion D _T by the number P of spectral estimates, the initial value d ₁ sample distortion value is first calculated (step 310). As will be described in detail below, only frequency coefficients from bands having a square (square) spectral estimate that is sufficiently larger than the final quantization distortion value d _w are encoded for transmission. Therefore, the permissible quantization value is used as a noise threshold for determining which coefficient is transmitted. Accordingly, module 74 performs an antilog operation on the log spectral estimate Log ₂ S _q ² (j) to form the square of the quantized spectral estimate S _q ² (j). Module 74 calculates the squared (squared) spectral estimate
If S _q ² (j) is equal to d ₁ less than or d _1, (referred to as "noisy sample" in this case) configuration frequency coefficients of the spectral estimates and those not encoded, tentatively assumed ( Step 312). Thus module 74, this structure coefficients to reflect the fact that not encoded, increasing the sample distortion value d _1. More specifically,
Module 74, all square equal to d ₁ is less than or d ₁ (2 squared) following equation sum D _NT spectral estimate (Step 314).

Module 74, then, by subtracting the sum from the collective allowable distortion D _T a (step 316) results, divided by the remaining number N of spectrum estimation value to calculate the adjusted sample distortion value d ₁ (step 318).

d ₁ = (D _T −D _NT ) / N (10) where N is the number of square (square) spectral estimates larger than the initial distortion value d ₁ . Since d ₁ can now be larger than its initial value d ₁ , module 74 compares each squared (squared) spectral estimate with the new d ₁ ,
It is determined whether any other coefficients are not coded (step 320). In that case, the estimator repeats the step of calculating the adjusted sample distortion value d ₁ (steps 322, 3
twenty four). Search when adjusted smaller to or equal to the additional square than the sample distortion value d ₁ (the square) spectral coefficients has run is terminated (step 320). The search also ends after the maximum number of searches allowed (step 322). The resulting sample distortion value d
₁ is then provided to bit rate comparator 76 (FIG. 7) (step 326).

Referring again to FIG. 7, the comparator 76 calculates the following equation based on the first sample distortion value d ₁ and the log spectrum estimated value Log ₂ S _q ² (j). , The provisional bit number r per frame is calculated.

Comparator 76 then compares the estimated number of bits to the maximum allowable number of bits per frame r _max . comparator 76 if r is smaller than the maximum number r _max is also notifies the Joule 78 to calculate the step size of the frequency coefficients based on the first sample distortion value d _1. However, if r exceeds the maximum number r _max , comparator 76 assumes that more distortion per estimate needs to be allowed to keep the number of bits less than r _max . Thus comparator 80 notifies the second distortion approximation module 80 to initiate an iterative search for a new distortion value d ₂ which gives a smaller bit rate than r _max.

Second Estimation of Allowable Distortion d ₂ Referring to FIG. 9, the approximation module 80 calculates Initially computing the distortion increment value D _i that satisfies (Step 41
0).

Strain increment value D _i is an estimate of the increment required in the first sample distortion value in order to reduce the bit rate below the maximum value R _max. Therefore, the approximation module calculates a new distortion value d ₂ that satisfies the following equation: Log ₂ d ₂ = D _i + Log ₂ d ₁ (13) (step 41)
2).

Approximation module, the same manner as described above, each of the square (the square) spectrum estimate S _q ² a (j) as compared to d _2, that defines which estimated value is equal to or smaller The strain than the strain , Which frequency coefficient is not coded (step 414). Based on this prediction, module 80 calculates the total number of bits r required for the frame as (Step 416).

Module 80 compares the rate r with a maximum value r _max ,
New distortion value d ₂ is determined whether or not give a small bit rate than the maximum value (step 418). If so, module 80 supplies the value d ₂ to module 78,
Quantization step size Q based on both values d ₂ and d ₁
The module 78 is instructed to calculate (j) (step 422). Otherwise, module 80, in an attempt to find a distortion d ₂ which gives a sufficiently low bit rate r, performs another iteration of the process (step 418,420～42
8). However, without finding such a distortion estimate,
If the maximum number of iterations has been attempted, the search is terminated and d ₂
Is supplied to the module 78 (steps 420 and 4).
twenty two).

Calculating Step Size and Classification Information from Distortion Estimates Referring to FIG. 10, module 78 combines weighted estimates of allowable sample distortion values d ₁ , d ₂ by weighting to satisfy Form the distortion d _w .

_{Log 2 d w = a log 2} d 2 + (1-a) log 2 d 1 (15) where the a is a constant weighting factor (step 51
0). (Note: If the value d ₂ was not calculated, weighted distortion d _w is placed equal to the first estimate d _1). Module 78 provides a class parameter c ′ for each spectral estimate S (j) based on the weighted distortion estimate d _w.
(J) is calculated according to the following equation (step 512).

The class parameter c '(j) is rounded upward to the nearest integer value in the range 0-8, and the class integer c (j)
Are formed (step 520). A class value of 0 indicates that all coefficients in the class should not be coded. Therefore, the class value c for each spectral estimate to indicate which coefficients to encode
(J) is supplied to the quantizer.

Class values greater than or equal to one are used to select the Huffman table used to encode the quantized coefficients. Accordingly, the class value c (j) for each spectral estimate is provided to Huffman encoder 47.

Module 78 then calculates, based on the value of the weighted distortion d _w and the value of the class parameter c ′ (j) for the estimate,
Band step size Q for each spectral estimate
(J) is calculated next (steps 514 to 518). More specifically, for spectral estimates with class values less than 7.5, the step size Q (j) is calculated to satisfy the following equation:

Where z is a constant offset value, for example -1.47156
(Step 516). Class parameter c '
For spectral coefficients where (j) is greater than or equal to 7.5, the step size is selected to satisfy the following equation (step 518).

Next, a step size Q (k) is derived for each frequency coefficient k in band j by interpolation of a band step size Q (j) for each band j. First
First, each band step size is scaled downward. For this, the spectral estimate S (j) for each band j is: 1) the sum of all parallel (squared) coefficients in the band, and 2) the 10 adjacent samples and blocks in the block. Recall that this was calculated as the larger of the two logical ORs of all coefficients (see FIGS. 2 (a) -2 (c)). Therefore, the selected spectral estimate is approximately equal to the total energy of all bands. However, the quantization step size for each coefficient should be chosen based on the average energy per coefficient in each band.

Therefore, the band step size Q (j) is determined by the number of coefficients (S,
It is scaled down by dividing by 10 or 20) according to the technique chosen for the calculation of (j). Next, the bit rate estimator 42 calculates the coefficient step size log ₂ Q
In order to calculate the logarithm of (k), linear interpolation is performed between the logarithmic band step sizes Log ₂ Q (j) (FIG. 11). Finally, the reciprocal of the coefficient step size is derived by:

1 / Q (k) − = (log ₂ Q (j)) ⁻¹ (19) Quantization and coding of frequency coefficient Referring to FIG. 1A, class integer c (j) and quantization step size Q (k) is supplied to the coefficient quantizer 44. The coefficient quantizer 44 has an associated inverse step size 1 /
A mid-trea that quantizes each frequency coefficient F (k) using Q (k) to generate an index I (k)
d) It is a quantizer. Exponent I (k) is encoded for transmission by the combined action of limit controller 45 and Huffman encoder 47.

The Huffman encoder 47 has a Huffman table containing Huffman codes of coefficients for each class c (j). The class integer c (j) is provided to a Huffman encoder 47 to select an appropriate Huffman table for the specified class.

The limit controller 49 generates a corresponding Huffman index I _h (k) that identifies an entry in the selected Huffman table in response to the index I (k). The Huffman encoder 47 then sends the selected Huffman code F _h (k) to the bit stream controller 20 for transmission to the far end.
Is transmitted.

Typically, limit controller 45 simply transmits index I (k) for use as Huffman coefficient I _h (k). However, the range of possible indices I (k) may exceed the input range of the corresponding Huffman table. For this reason, the limit controller 45 has a maximum exponent value and a minimum exponent value for each class c (j). The limit controller 45 determines each index I
Compare (k) with these maximum and minimum exponent values. I
If (k) exceeds either the maximum or minimum exponent value,
The limit controller 45 clips I (k) to be equal to the respective maximum and minimum values, and uses this clipped exponent as the corresponding Huffman exponent.
Supply 47.

The limit controller 45 holds a running tally of the number of bits necessary for transmitting the Huffman code for each frame. More specifically, the limit controller has a table of the number of bits corresponding to each Huffman table of the Huffman encoder 47. Each entry in the bit number table represents the number of bits of the corresponding Huffman code stored in the Huffman table of encoder 47. Therefore, each Huffman index I _h generated by the limit controller 45
For (k), the limit controller determines the corresponding Huffman code F identified by the Huffman index I _h (k)
_The Huffman exponent is provided internally to a bit number table to determine the number of bits required to transmit _h (k). The number of bits is then added to the running tally. If the running tally exceeds the maximum number of allowed bits, the limit controller 45 ignores the remaining index I (k). The limit controller 45 then informs the far-end receiver that the allowed number of bits in the frame has been reached and that the remaining coefficients are not encoded for transmission.
A unique Huffman code for identifying a unique Huffman code is prepared.

To transmit this unique Huffman code, the limit controller must allocate multiple bits. For this reason, the limit controller discards the most recent Huffman code and recalculates the running tally to determine if enough bits are available to carry a unique Huffman code. If not, the limit controller discards the most recent Huffman code until a sufficient number of bits are allocated for the transmission of a unique Huffman code.

Reconstruction of Microphone Signal at Far End Referring again to FIG. 1 (b), the audio decoder 28 at the far end reconstructs a microphone signal from a set of encoded signals. More specifically, the Huffman decoder 25
Decoding Huffman codes S _h a (j) to reconstruct the encoded logarithmic spectrum estimate S _e (j). The decoder 27 (the decoder 41 of the audio encoder 16 in FIG. 1A)
) Further decodes the encoded logarithmic spectral estimate to produce a quantized spectral estimate Log ₂
Reconstruct S _q ² (j).

The log spectral estimate Log _j S _q ² (j) and the received signal index A _i (F) are used again to derive the step size Q (k) and class c (j) performed by the near end bit rate estimator 42. It is provided to the bit rate estimator 46 for performing. The derived class information c (j) is supplied to the Huffman decoder 47 for decoding the Huffman code F _h (k).
The output of the Huffman decoder 47 is output to a coefficient reconstruction module 48.
, And reconstructs the original coefficient F _q (k) based on the derived quantization step size Q (k). The bit rate estimator 46 further performs coefficient filling (co
It supplies class information to the efficient fill-in) module 50 and informs the coefficient filling module 50 which coefficients were not coded for transmission. Module 50 then performs a reconstructed log spectral estimate Log ₂ S _q ² (j)
Is used to estimate the missing coefficient.

Finally, the decoded coefficient F _q (k) and the estimated coefficient F _e (k) are supplied to the signal
52, again the amount of time back to input coefficient values is de normalize the time-domain signal using the reconstructed normalization gain g _q.

More specifically, the inverse DCT module 51 merges the decoded coefficients with the estimated coefficients F _q (k), F _e (k), and converts the resulting frequency coefficient values into the time domain signal m ′ (n ). The inverse normalization module 53 returns the time domain signal m '(n) to the original scale. The generated microphone signal m (n) is supplied to an overlap decoder 55, which removes redundant samples introduced by the window module 32 of the audio encoder 16 (FIG. 1 (a)). Signal conditioner 57
Converts the obtained microphone signal into an analog signal L (t) for driving the speaker 29.

Estimation of Uncoded Coefficients As mentioned above, in the signal spectral region having relatively high energy levels of energy, the human ear is less likely to notice distortion of the audio signal. Therefore, the bit rate estimator 42 (FIG. 1) adjusts the quantization step size so that the coefficients at the low energy level are fine and the coefficients at the high energy level are coarse, respectively. In sharp contrast to this approach, the bit rate estimator simply discards the coefficients at very low energy levels.

The absence of these coefficients will result in a significant amount of audio artifacts. Therefore, the coefficient filling module 50 creates a coefficient estimate for each discarded coefficient from the spectral estimate. Factor module during this time, considering the level of the signal index A _i for frame coefficients that are not coded are present. For example, if the signal index is low (indicating that the frame mainly consists of background noise), the filling module estimates that the missing coefficients are representative of background noise. Therefore, the filling module
The coefficient is created mainly using the background noise in the frame as a guide. However, if the signal index is high (indicating that the frame mainly consists of the speech signal), the noise filling module assumes that the missing coefficients are representative of the speech signal. Thus, the filling module creates coefficient estimates from the spectral estimates corresponding to the frequency bands that mainly include the frequency of the lost coefficients.

Referring to FIG. 12, coefficient filling module 50 (FIG. 1)
(B)) is a coefficient estimator module for each band.
It has 82. Each estimator module 82 includes a noise floor module 84 for approximating the amount of background noise in frequency band j for each frame F. The noise estimate S _n (j, F) is derived by comparing the log spectral designation of the current frame Log ₂ S _q ² (j) with a noise estimate derived from the spectral estimation of the previous frame. The adder 91 adds the logarithmic spectrum estimate to the logarithmic gain log ₂ g _q ² to denormalize the logarithmic spectrum estimate. The denormalized estimate S _u (j, F) is given by S
_u (j, F) is provided to a comparator 99 which compares the noise estimate _Sn (j, F-1) calculated for the previous frame F-1. (S _n has been initialized to zero for the first frame). S _u (j, F) is the previous noise estimate S _n (j, F−1)
If so, the noise estimate for the current frame F is calculated by:

S _n (j, F) = S _n (j, F−1) + t _r (20) where _tr is the rise time constant given by table 100. More particularly, table 100 may donate unambiguous t _r for each value of the signal index A _i (F). (For example, for A _i (F) = 0, a relatively long time constant is selected to give a long time constant. A _i (F)
Increases, the selected time constant decreases).

If S _u (j, F) is smaller than the previous noise estimate S _n (j, F−1), the noise of the current frame is calculated by the following equation.

S _n (j, F) = S _n (j, F−1) + t _f (21) where t _f is a rising time constant given by the table 102. Like the table 100, the table 102 has a unique time constant t _f for each index A _i (F) value.
give. Adder 93 subtracts the logarithmic gain log ₂ g ² _q to obtain the resulting noise estimate S _n (j,
F). The output of 93 is provided to a logarithmic inverter 64 that calculates the antilog according to the following equation and provides a normalized noise estimate S _nn (j, F):

For each frame, the normalized band noise is S
_nn (j, F) and a spectral estimate _Sq (j, F) is a weighting function that produces a weighted sum of these binary values.
Supplied to 86. The weighting function 86 is the signal index A _i
A first table 88 including a noise weighting coefficient C _n (F) for each value of (F) is included. Similarly, the second table 90
For each index A _i (F), a speech weighting index C
_a (F) is included. Table 88 shows the audio index A
_In response to the current value of _i (F), a corresponding noise weighting factor C _n (F) is provided to multiplier 92. The multiplier 92 calculates C _n
Calculate the product of (F) and S _nn (j, F) to form a weighted noise value S _nw (j, F). Similarly, table 90
Calculate the weighted speech value S _vw (j, F) (S _vw = S _q
(J, F) C _a ). The weighted values are provided to a former 98 which calculates an estimate of the weight according to the following equation:

W = (S _vw + S _nw ) ² (23) The weighting coefficients C _a and C _n stored in the tables 90 and 88 are
It is related by C _a = 1−C _n . Where the value of C _n ranges from 0 (for A _i = 7) to 1 (for A _i = 0)
In the range up to. Thus, during silence, the noise estimate has more weight and the spectral index is given more weight as the audio index increases.

This weighted estimate is provided to the signal shaper 52 for use in calculating the estimated frequency coefficient F _e (k) for each of the ten unsigned frequency coefficients corresponding to the spectral estimate. Is done.

More specifically, the weighted estimate W is used to control the “fill-in” level for each missing frequency coefficient (ie, one with 0 class). Filling consists of inserting the result F _e instead of scaling the outputs from the (uniform distribution with) a random number generator, lost frequency coefficients. The following equation is used to generate a fill-in for each missing frequency coefficient.

Where noise is the output (or coefficient y samples) from the random number generator at a given instantaneous point,
The range of the random number generator is twice the value n. e is a constant, for example, 3. A new noise value is generated for each missing frequency coefficient.

Additions, deletions, or other modifications of the preferred feature embodiments of the present invention will be apparent to those skilled in the art and are encompassed by the following claims.

Continuation of the front page (56) References JP-A-56-66929 (JP, A) JP-A-64-24513 (JP, A) JP-A-63-151269 (JP, A) JP-A-56-46300 (JP, A) , A) JP-B 60-37658 (JP, B1) JP-B 2-52280 (JP, B2) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 19/00

Claims (37)

(57) [Claims] 1. A method for allocating transmission bits used to transmit samples of a digital signal, the method comprising: collective allowable quantization representing a collective allowable quantization distortion error for a frame comprising a plurality of samples of the digital signal. Selecting a distortion value; selecting a set of samples from the frame of samples such that each of the set of samples is greater than a noise threshold; and for each sample of the set, Calculating a sample quantization distortion value representing an allowable quantization distortion error of the sample;
Sample quantizing the set of all samples such that the sum of the distortion values is approximately equal to the collective allowable quantization distortion value; and for each sample of the set, the corresponding quantum of the samples A quantization step size that gives a quantization distortion error substantially equal to the quantization distortion value, and quantizing the sample using the quantization step size. 2. The method according to claim 1, wherein the step of selecting a quantization step size is based at least in part on a difference between the sample and a corresponding quantization distortion value of the sample. 2. The method according to claim 1, further comprising the step of calculating. 3. The digital signal includes a noise component,
The selection of the collective allowable quantization distortion comprises creating a signal index that represents the signal magnitude of the signal component relative to the noise magnitude of the noise component for at least one sample of the frame, and based on the signal index. 2. The method according to claim 1, further comprising the steps of: selecting the collective allowable quantization distortion. 4. The method of claim 1, wherein calculating the sample quantization distortion value comprises dividing the collective allowable quantization distortion value by a number of samples in a frame of the plurality of samples to form a first sample distortion value. The allocation method according to claim 1. 5. The method according to claim 1, wherein the selection of a set of samples is such that the trial set of samples of the digital signal is more trial than a noise threshold defined at least in part by a value of the first sample distortion value. Selecting each sample of the set to be large, wherein calculating the sample threshold comprises:
5. The method of claim 4, further comprising adjusting the first sample distortion value by a value determined by a difference between the first sample distortion value and at least one sample removed from the trial set. The assignment method described. 6. The trial selection of a set of samples and the adjustment of the first sample distortion value include the steps of: a) adjusting the first sample distortion value to a value less than a current value of the first sample distortion value; Identifying the noisy samples in the set; b) removing any noisy samples from the trial set, if any; c) determining if the noisy samples are removed. The first sample distortion value also increases the first sample distortion value by an amount determined by a difference between at least one of the noisy samples, the trial sample value being greater than the adjusted first sample distortion. A method comprising: repeating steps a), b) and c) of the set of additional noisy samples until the adjusted first sample distortion value is reached such that step a) is no longer specified. 6. The allocation method according to claim 5, wherein: 7. The method of claim 6, further comprising the step of aborting said adjustment if steps a), b) and c) are repeated a certain maximum number of times. 8. After the adjustment is complete, estimate the number of bits for the amount of bits required to transmit the trial set of all samples, and compare the estimated number of bits to a maximum number of bits. And further comprising, if the estimated number of bits is less than or equal to the maximum number of bits, selecting a final noise threshold based on the adjusted first sample distortion value. 9. The allocation method according to claim 7, wherein: 9. A second sample distortion value is generated if the estimated number of bits exceeds the maximum number of bits; d) a plurality of samples of the digital signal of a second trial set. And selecting each of the plurality of samples to have a value greater than the second sample distortion value; e) the number of bits required to transmit all the samples of the second trial set. E) comparing said estimated number of bits to said maximum number of bits; g) said second number by an amount determined by said selection if said estimated number of bits is greater than said maximum number of bits. 9. The method according to claim 8, further comprising the steps of: increasing a sample distortion value. 10. Repeating steps d) to g) until a second sample distortion value is reached as determined by step g that said estimated number of bits is less than or equal to said maximum number of bits. 10. The method of claim 9, further comprising the step of: 11. The sampled distortion value is calculated from the adjusted first sampled distortion value and the second sampled distortion value, a final set of digital signal samples is selected, and the final set of a plurality of samples is selected. 11. The method according to claim 10, further comprising the step of: having each value of each of said first and second threshold values be greater than a certain final threshold value determined by a corresponding sample distortion value of said sample. 12. A method for communicating a digital signal including a noise component and a signal component, comprising: generating an estimated signal having a smaller number of samples than the digital signal representing the digital signal; Creating a signal index representing the magnitude of the signal component signal for at least one sample of the estimated signal; selecting a sample of the digital signal having a sufficiently large signal component based on the signal index; A communication method for transmitting the selected sample of the digital signal, transmitting the sample of the digital estimation signal, and reconstructing the digital signal from the transmitted selected sample and the estimation signal. 13. The digital signal is a frequency domain audio signal representing audio information to be transmitted, and each sample of the estimated signal is a spectrum estimate of a frequency domain signal of a corresponding frequency band. Reconstructing a domain audio signal, generating a random number for each unselected sample of the frequency domain audio signal, creating a noise estimate of the value of the noise component of the unselected sample for at least one of the spectral estimates Generating a scaling factor based on the noise estimate; and scaling the random number according to the scaling factor to form a reconstructed sample representing the unselected sample. The communication method described. 14. The estimation signal and the frequency domain signal each include a series of frames, each frame representing the audio information over a specified time window;
A first noise estimate of a current frame comprising: for a certain previous frame of the estimated signal, creating an initial noise estimate from at least one spectral estimate representing a frequency band of the previous frame; based on, select the rise time constant t _r and fall time constant t _f, to form a upper threshold by adding the upstanding uplink time constant outermost first noise estimate, the outermost first the upstanding edge time constant Subtracting from the noise estimate to form a lower threshold, comparing a current spectral estimate of the current frame, representing the frequency band, with the upper and lower thresholds, wherein the current spectral estimate is the threshold If the current noise estimate is equal to the current spectral estimate, and if the current spectral estimate is less than the lower threshold, Subtracting the noise estimate by the falling time constant to form an estimate of the current noise; if the current spectral estimate is greater than the upper threshold, reducing the first noise estimate by the rising time constant 14. The communication method according to claim 13, comprising the steps of incrementing to form the current noise estimate. 15. The method of claim 14, wherein generating a scaling factor includes generating a weighted sum of the current noise estimate and the current spectral estimate. 16. The generation of the weighted sum comprises: selecting a noise factor based on the value of the signal index; selecting a speech factor based on the value of the signal index; 16. The method according to claim 15, further comprising: multiplying a noise coefficient, multiplying the current spectrum estimation value by the audio coefficient, adding the multiplication result, and forming the scaling factor from a result of the addition. Communication method. 17. An encoding device for allocating transmission bits used for transmitting a plurality of samples of a digital signal, the apparatus comprising: Means for selecting a total allowable quantization value to represent; and means for selecting a set of samples from the frame of samples such that each of the set of samples is greater than a noise threshold. And for each of the set of samples, a sample quantization distortion value representing an allowable quantization distortion error of the sample, and for each of the set of samples, all sample quantizations of the set of all samples. Means for calculating the sum of the distortion values to be approximately equal to the collective allowable quantization distortion; and, for each sample of the set, the sample Encoding means for selecting a quantization step size that gives a quantization distortion error approximately equal to the corresponding quantization distortion value of the sample, and means for quantizing the samples using the quantization step size. . 18. The method according to claim 18, wherein said means for selecting a quantization step size comprises: for each sample to be transmitted, a quantization step size that is at least partially equal to a difference between said sample and a corresponding quantization distortion value of said sample. 18. The encoding apparatus according to claim 17, further comprising means for performing calculation based on the target. 19. The digital signal includes a noise component and a signal component, wherein the means for selecting a collective allowable quantization distortion comprises: a signal magnitude of the signal component with respect to a noise magnitude of the noise component. At least one of the frames
18. The encoding apparatus according to claim 17, further comprising: means for generating a signal index representing one sample; and means for selecting the collective allowable quantization distortion based on the signal index. 20. The means for calculating the sample quantization distortion value, wherein the collective allowable quantization distortion value is divided by a number of samples in a frame of the plurality of samples to obtain a first sample distortion value. Claims including means for forming
Item 18. The encoding device according to Item 17, 21. The means for selecting a set of the plurality of samples includes: a trial set of a plurality of samples of the digital signal; Means for selecting to be greater than a noise threshold at least partially determined by a value of the first sample distortion value, wherein the means for calculating the sample quantization distortion value comprises: 21. The encoding method according to claim 20, further comprising: means for adjusting the first sample distortion value by an amount determined by a difference between the first sample distortion value and at least one sample excluded from the trial set. . 22. The means for selecting a trial set of samples and the means for adjusting the first sample distortion value, wherein each of the means has a value less than a current value of the first sample distortion value. Means for identifying a trial set of noisy samples; means for removing the noisy samples, if any, from the trial set; and, if the noisy samples are removed, the first sample distortion value. Means for increasing the first sample distortion value by an amount determined by a difference between at least one of the noisy samples and at least one of the additional noisy samples of the trial set rather than the adjusted first sample distortion value. Until the adjusted first sample distortion value is reached such that it does not increase
22. The encoding apparatus according to claim 21, further comprising: means for removing a noisy sample and adjusting the first sample distortion value. 23. The encoding apparatus according to claim 22, further comprising means for terminating said adjustment when said first sample distortion value has been adjusted a maximum number of times. 24. A means for estimating the number of bits for the amount of bits necessary to tradition all the samples of the trial set after the adjustment is completed; and estimating the estimated number of bits to a maximum number of bits. Means for comparing the number of bits with the number of bits and selecting a final noise threshold based on the adjusted first sample distortion value if the estimated number of bits is less than or equal to the maximum number of bits. 24. The encoding device according to claim 23, further comprising: 25. A means for producing a second sample distortion value if the estimated number of bits exceeds the maximum number of bits, and a second trial set of samples of the digital signal. Means for each of said second trial set of samples to have a value greater than said second sample distortion value; and transmitting said second trial set of all samples. Means for estimating the number of bits required to perform the operation, means for comparing the estimated number of bits with the maximum number of bits, and if the estimated number of bits is greater than the maximum number of bits, 25. The encoding apparatus according to claim 24, further comprising: means for increasing the second sample distortion value by a determined amount. 26. Iteratively adjusting the second sample distortion value;
Re-selecting the second set of trial samples,
Further comprising means for estimating all the samples of the trial set of E., until the estimated number of bits reaches a second sample distortion value such that the estimated number of bits is less than or equal to the maximum number of bits. 26. The encoding device according to item 25. 27. A means for calculating the sample distortion value from the adjusted first sample distortion value and the second sample distortion value; and a final set of a plurality of samples of the digital signal, 27. The encoding apparatus according to claim 26, further comprising: means for selecting each of the plurality of samples to have a value greater than a last threshold determined by a corresponding sample distortion value of the sample. 28. An apparatus for transmitting a digital signal including a noise component and a signal component, comprising: means for generating an estimated signal having a smaller number of samples than the digital signal representing the digital signal; Means for generating, for at least one sample, a signal index representing the ratio of the value of the signal component to the value of the noise component; and determining a sample of the digital signal having a signal component that is sufficiently large based on the signal index. Means for selecting; means for transmitting selected samples of the digital signal; means for transmitting the samples of the digital estimated signal; means for reconstructing the digital signal from the selected transmitted and estimated samples. An encoding device comprising: 29. The digital signal is a frequency domain audio signal representing audio information to be transmitted, and each sample of the estimated signal is a spectrum estimate of a frequency domain signal in a corresponding frequency band. Means for reconstructing the audio signal in the frequency domain, means for generating a random number of each unselected sample of the audio signal in the frequency domain, and Means for generating from the one or more spectral estimates; means for generating a scaling factor based on the noise estimate; and a reconstructed sample representing the unselected sample by scaling the random number according to the scaling factor. 29. The encoding apparatus according to claim 28, further comprising: means for generating a signal. 30. The estimation signal and the frequency-domain audio signal each include a series of frames, each of which represents the audio information over a specified time window, and includes one for the current frame. Means for generating a next noise estimate, for a previous frame of the estimated signal, means for generating an initial noise estimate from at least one spectral estimate representing a frequency band of the previous frame; based on the value of the exponent, and means for selecting a rise time constant t _r and fall time constant t _f, means for forming the threshold value of the upper by adding the upstanding uplink time constant outermost first noise estimate, the Means for subtracting the fall time constant from the initial noise estimate to form a lower threshold; and applying a current spectral estimate representing the frequency band in the current frame to the upper Means for comparing the current noise estimate with the current spectrum estimate when the current spectrum estimate is in the middle of the threshold value; and Means for reducing the initial noise estimate by the fall time constant when the value is lower than the lower threshold to form the current noise estimate; and wherein the current spectrum estimate is greater than the upper threshold. Means for incrementing said first noise estimate by said rise time constant to form said current noise estimate when said first noise estimate is also higher. 31. The encoding apparatus according to claim 30, wherein said means for generating a scaling factor includes means for generating a weighted sum of said current noise estimate and said current spectral estimate. 32. The means for generating the weighted sum comprises: means for selecting a noise coefficient based on a value of the signal index; means for selecting a voice coefficient based on a value of the signal coefficient; Means for multiplying a current noise estimate by the noise coefficient; means for multiplying the current spectrum estimate by the speech coefficient; means for adding the multiplication result; and forming the scaling factor from the addition result. 32. The encoding device according to claim 31, comprising: means. 33. An estimating method for estimating the energy of a speech component for each of a series of frames of a digital signal, comprising: for a first frequency band of a first frame, a noise energy in the first frequency band of the first frame; Creating a first frame noise estimate that represents the first frame noise estimate and adding a small increment to the first frame noise estimate when the signal energy value is greater than the first frame noise estimate; Forming a noise estimate, and if the signal energy value is less than the first frame noise estimate, subtracting a large decrement from the first frame noise estimate to form a second frame noise estimate Subtracting the second frame noise estimate from the value of the code energy in the first frequency band, An estimation method comprising: forming a speech estimate that represents energy of a speech component of a first frequency band. 34. For each of the plurality of frequency bands of the second frame to create an estimate of speech for each speech estimate by subtracting the voice estimate from the threshold S _T, bandwidth estimation value S to form a _out, selects the maximum bandwidth estimate S _out of the maximum bandwidth estimation value S _max, based on the maximum bandwidth estimation value S _max, the frame speech representing the energy of the entire band of the speech component of the second frame 34. The estimation method according to claim 33, further comprising each step of creating an estimated value. 35. The method according to claim 35, wherein generating the frame estimate comprises the step of setting the frame speech estimate equal to zero if the maximum bandwidth estimate S _max is less than or equal to zero.
The estimation method according to paragraph 33. 36. The sample of each frame is normalized by a frame normalization factor g, and the creation of a signal energy value representing the energy of the digital signal is such that the denormalized signal is the same as said first frame noise estimate. 34. The method of claim 33, comprising denormalizing the second frame of the digital signal to be on a scale. 37. A method for generating a frame speech estimate comprising: multiplying said maximum bandwidth estimate S _max by a weighting factor to form a _skateled value S ′ _max , representing a scaled value of a normalized gain g. create an attenuated gain g _o, if said maximum bandwidth estimation value S _max is greater than 0, S _'max and g _o
37. The method according to claim 36, further comprising the step of: selecting the larger one of the above as the frame sound estimation value.