JPS6348080B2 - - Google Patents

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention uses only effective word components from the audio signal in order to subject audio information converted to an electrical signal to predetermined processing in a speech processing device such as a speech response or speech recognition device. This invention relates to an automatic concentration method for audio information when extracting audio information and storing it on an output recording medium.

[Technical background of the invention]

For example, in a voice response device, various fragmentary phrases are stored in a storage device in advance, and when making a voice response, some of the phrases are selected and sequentially reproduced in a predetermined order. Meaningful response sentences were appropriately synthesized and output. In this case, if there are unnecessary silent parts before and after each word to be stored in the storage device, not only will the information storage efficiency in the storage device be reduced, but also an unnatural response sentence may result during synthetic playback. If necessary, silent parts can be added as appropriate during synthetic reproduction, so when storing words in a storage device, it is desirable to eliminate unnecessary silent parts as much as possible and store only the words.

Conventionally, when storing phrase information in a storage device in this type of device, various phrases are uttered at appropriate intervals in advance and recorded on a recording medium such as a magnetic tape, and the operator records the phrases while playing back the recorded phrases. The beginning and end of words are determined, and additional information such as an address and a search identification code necessary for reading a word is added thereto, and the words are recorded again on another recording medium.

That is, as shown schematically in Figure 1, which shows the speech waveform before decentralization, there are silent sections S before and after the words W1 and W2.
1, S2, and S3 exist, but this silent section S1
An operator listens to and recognizes ~S3, determines the beginning and end of a phrase, and performs a collection of phrases. For this reason, it takes a long time to concentrate, and the operator's physical
The mental burden was also considerable.

On the other hand, in order to easily perform biasing, for example, a speaker utters a scheduled word based on an operator's instructions, and the operator records only this word on a recording medium while processing the words before and after the word (additional information It is conceivable to add however,
In this case, the concentration work itself is simplified once again, but the vocalization and the operator must be present at the same time during concentration, in which case a recording device for editing must be present, and surrounding noise etc. must be present at the same time. This increases additional conditions, such as the need to be in a location such as a recording studio with limited availability, and as a result, editing becomes even more complicated.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide a method for automatically collecting speech information in a speech processing device, which makes it possible to automatically collect word information without any human intervention when editing speech information.

[Summary of the invention]

The present invention arbitrarily sets a reference level and a reference time in advance to identify a silent section, compares these with the input audio signal, and selects only the parts with a high level between the certified silent sections and the word/phrase speech. The feature is that only this part is recognized and stored on the output medium with additional information attached.

[Embodiments of the invention]

FIG. 2 shows the configuration of an apparatus to which an embodiment of the present invention is applied, and FIG. 3 shows the processing flow of the embodiment.

In FIG. 2, reference numeral 1 denotes an input terminal into which the audio signal to be edited is input, and the audio information to be concentrated and stored is inputted in the form of an electric signal and at a certain time interval or more for each word. 2 is an A/D (analog/digital) converter that converts the analog audio signal input to the input terminal 1 into digital information;
Reference numeral 3 denotes an input device for setting and inputting a reference level and reference time for recognizing a silent section.

Here, the reference level is the lowest level that is considered to be non-silent, and is information for distinguishing speech from noise etc. in the input audio signal, and anything below this reference level is considered to be no speech, that is, a silent state. In addition, the reference time is the minimum duration of silence that is considered to be in a word and phrase, and is information for distinguishing between a short period of silence within a word and a silence between words. When the condition persists,
This is regarded as a silent interval between words. Therefore,
This reference time is set based on the time interval between words of the input audio signal described above.

4 is a concentrated processing device having a buffer memory 4a, which is configured using, for example, an electronic computer,
Based on the reference level and reference time input and set by the input device 3 described above, the digitized input audio signal input from the A/D converter 2 is discriminated between silent intervals between words and phrases, and only the word portions are discriminated. Extract. Reference numeral 5 denotes an output recording device that stores audio information of words extracted by the polarization processing device 4 on an output recording medium.

The operation in such a configuration will be explained with reference to FIG.

Prior to the actual concentration processing, the aforementioned reference level and reference time settings are entered through the input device 3 as a preliminary process (processing P1), and this information is stored in the buffer memory in the concentration processing device 4. 4
a (processing P2). Next, when the original editing process is started, the input audio signal to be edited is first input from the input terminal 1 (processing P3), A/D converted by the A/D converter 2, and encoded into digital information. (Processing P4). The input audio signal digitized in this way is compared with the reference level held in the buffer memory 4a within the concentration processing device 4 (processing P5), and if it is equal to or higher than the reference level, it is determined whether flag A is set or not. Check (processing P6), and if it is not set, give the input audio signal to the output recording device 5 and store it in the output recording medium (processing
P7), then the next audio signal input processing (processing P3)
to move to.

The input audio signal is compared to a reference level (processed
P5), if the input audio signal is in a silent state below the reference level, check whether flag B is set (processing P8), and if not set, check whether the duration is longer than the reference time or not. is determined (processing P9). If the silence duration time is less than the reference time, the input audio signal is stored as silence section data in a buffer memory, etc., and at the same time, a flag A indicating that the silence section data is saved is set (processing 10), and then the next step is performed. Shift to audio signal input processing (processing P3).

As a result of determining the silence duration (processing P9), if the silence duration is equal to or greater than the reference time, this means that a silence interval between words has been detected, so the measurement data for the silence duration is initialized. , clears the saved silent section data and resets flag A (processing P11), and adds an address, search identification code, etc. to the series of audio data that has been output to the output recording medium as one word. After adding information (processing P12) and performing initialization processing for the next audio data processing (processing P13), flag B is set to skip input silent data if the silent state continues. (processing P14), and moves to the next audio signal input processing (processing P3). Therefore, if silent data is input in this state,
After level recognition processing (processing P5), it is determined whether flag B is set (processing P8), and since flag B is set, the process immediately moves to the next audio signal input processing (processing P3).

As a result of input audio signal level judgment (processing P5),
If it is determined that the input audio signal is above the reference level and flag A is set (processing P6), the saved silent section data is a short silent section in a word, so it is left as is. Store it in the output recording medium (processing P15), then clear the saved silent section data, reset flag A, initialize the measurement time (silent duration) (processing P16), and then output and record the input audio data. The audio signal is stored in the medium (processing P7), and the process moves on to capturing the next audio signal (processing P3).

In this way, only word information is automatically extracted from the input audio signal, and the required information is added and stored on the output recording medium. Therefore, the task of collecting and storing audio information is dramatically simplified.

Note that the present invention is not limited to the embodiments described above and shown in the drawings, but can be implemented with various modifications without changing the gist thereof.

For example, in the above embodiment, the audio signal input to the input terminal 1 may be a signal reproduced from a recording medium or a signal input directly via a microphone, etc. Of course, the edited audio data may be a signal that is input as a voice response. Since word discrimination is necessary not only for use as stored voice data in , but also for voice processing such as voice recognition, the present invention can be applied to input processing for such voice processing.

In addition, the processing algorithm is not limited to the one shown in Figure 3, but the point is that silence is recognized based on a preset reference level, and only when this silence continues for longer than a preset reference time, the silence is detected. It is sufficient to store a series of audio data up to the start of the section as one word and to exclude the silent section as unnecessary data.

〔Effect of the invention〕

According to the present invention, it is possible to provide an automatic editing method of audio information in an audio processing device that can realize automatic editing processing of phrase information when editing audio information.

[Brief explanation of the drawing]

FIG. 1 is a schematic audio waveform diagram showing an example of the audio information waveform to be processed, FIG. 2 is a block diagram showing the configuration of an automatic editing device to which an embodiment of the present invention is applied, and FIG. 3 is a diagram of the present invention. 2 is a flowchart showing the processing flow of one embodiment. 1...Input terminal, 2...A/D converter, 3...
...Input device, 4...Editing processing device, 4a...Buffer memory, 5...Output recording device.

Claims (1)

[Claims] 1. In extracting only effective components from an audio signal and storing them in an output recording medium in order to subject the audio information converted to an electrical signal to predetermined processing in the audio processing device, silence preset from the outside is provided. Based on the reference level for determining the state and the reference time for determining the silent state, the first condition is that the input audio signal is less than the reference level, and a flag B for skipping the input silent data is set. If so, the next audio signal is input, and the second condition is that the input audio signal is less than the reference level, flag B for skipping input silent data is reset, and the duration of the input audio signal is less than the reference level. is less than the reference time, the input audio signal is stored in the buffer memory as silent section data, and a flag A indicating that silent section data is stored in the buffer memory is set, and then the input of the next audio signal is started. However, as a third condition, if the input audio signal is less than the reference level and the duration of less than the reference level is longer than the reference time, the measurement data of the duration of less than the reference level is initialized and the data is stored in the buffer memory. Clears the silent section data stored in the buffer memory, resets flag A indicating that silent section data is stored in the buffer memory, and sends an address to the output recording medium in order to convert the stored series of audio data into one word. Flag B for adding additional information such as a search identification code, performing a termination process in preparation for the next audio data processing, and skipping input silent data when the silent state continues thereafter.
After setting, the next audio signal is input, and as a fourth condition, a flag A is set indicating that the input audio signal is above the reference level and that silent section data is stored in the buffer memory. If the silent interval data is stored in the output recording medium, the silent interval data already stored in the buffer memory is cleared, and the flag A indicating that the silent interval data is stored in the buffer memory is reset. After initializing the measurement data of the duration below the reference level and storing the audio signal in the output recording medium, the input of the next audio signal is started, and as a fifth condition, the input audio signal is at the reference level or higher and the buffer memory is When a flag A indicating that silent section data is stored is reset, the audio signal is stored in the output recording medium and then processing is performed to input the next audio signal. A method for automatic concentration of audio information in an audio processing device.