EP2394271B1 - Method for separating signal paths and use for improving speech using electric larynx - Google Patents

Description

The invention is a method for improving the speech quality of an electro-laryngeal (EL) speaker, wherein the speech signal of the speaker is digitized by suitable means. By suitable means is meant, for example, a microphone with associated analog / digital converter, a telephone or other methods using electronic equipment.

An EL is an artificial spare voice device, for example, for patients who have had their larynx surgically removed. The EL is attached to the underside of the jaw; a tone generator with a certain frequency makes the air in the oral cavity vibrate over the soft tissues on the underside of the jaw. These vibrations are then modulated by the articulation organs so that speaking becomes possible. However, since the tone generator usually only works with one frequency, the voice sounds monotonous and unnatural, or "robotic".

Furthermore, it is disadvantageous that the vibration of the EL disturbs the perception of the speech or even drowned out, because only a part of the sound is articulated in the oral cavity. The parts emerging directly from the device or at the transition point on the neck overlay the articulated parts and reduce the intelligibility. This is particularly the case with speakers who have received radiation therapy in the neck area, which stiffens the tissue structure. Various methods have therefore been developed which are intended to amplify the useful signal - ie the articulated vibrations - in relation to the interference signal - ie the direct sound or the unmodulated vibration of the EL.

These methods are predominantly used in situations in which the listener is not directly exposed to the radiated sound, but electronic mediators are used, for example, when making a call, with sound recordings or generally when speaking through microphone and amplifier.

In the US Pat. No. 6,359,988 B1 an EL vocal signal is subjected to cepstrum analysis and superimposed with the speech of a normal speaker, whereby the pitch change of the EL speaker can be made more natural; At the same time, this also suppresses the proportion of direct sound emitted by the signal. The disadvantage of this solution is, above all, that for every statement of an EL spokesman at the same time the same statement of a healthy (ie without EL speaking) speaker is needed, which is practically impossible to achieve.

Another solution shows the US 6,975,984 B2 in which a solution for improving an EL speech signal in the telephony will be described. In this case, the speech signal is processed in a digital signal processor such that the buzzing background noise of the EL is detected and removed from the speech signal. The speech signal is divided into a voiced and an unvoiced component and processed separately. The voiced part is Fourier-transformed in blocks, frequency-filtered (fundamental frequency and harmonics are reused), inverse transformed and subsequently subtracted from the entire original signal. What remains is the unvoiced portion of the original signal. Alternatively, it is also proposed to filter the voiced portion over lowpass, completely filter out in case of recognition of a speech break and to superimpose the unvoiced portion afterwards.

The document " Enhancement of Electrolaryngeal Speech by Adaptive Filtering "by Carol Y. Espy-Wilson et al. (JSLHR, 41: 1253-1264, 1998 ) describes a method for improving the speech quality of an EL speaker. The basic noise of the EL is adapted by means of adaptive filtering to the disturbed by the EL fundamental noise speech signal (or the articulated to speech EL fundamental noise); in a further step, the signals are subtracted from each other. What remains is an error signal which is used to control and adapt the filter parameters with the aim of minimizing the error signal. The error signal in the present method is the speech signal freed from EL fundamental noise. The assumption is that although the interference signal in the speech signal is correlated with the EL fundamental noise, the speech signal of interest is independent of the other signals, that is to say that the disturbing background noise and the speech signal originate from different sources.

The document " Enhancement of Electrolarynx Speech Based on Auditory Masking "by Hanjun Liu et al (IEEE Transactions on Biomedical Engineering, 53 (5): 865-874, 2006 ) describes a subtraction algorithm for improving the signal of an EL speaker, in particular with respect to environmental noise.

Unlike other methods that provide fixed subtraction parameters, this algorithm adapts the subtraction parameters in the frequency domain based on auditory masking. It is assumed that speech and background noise are uncorrelated and therefore the background noise can be estimated and subtracted from the signal in the frequency domain.

These solutions have in common that methods are used based on a model, according to which speech and noise (ie ambient noise, but also the background noise of the EL) are statistically independent, or uncorrelated.

Based on this assumption, the implementation of the methods mentioned is carried out in a very complex manner. Attempting to suppress direct sound with an (adaptive) notch filter also reduces the quality of the speech signal, which then sounds like a whisper; Speech signal and noise are on the same harmonics.

The US 2005/0004604 A1 describes a laryngeal solution in which a sounder and a microphone are placed directly in front of a user's mouth, the sounder delivering a low volume sound and recording the signal for further processing via the microphone. In further processing, the signal is essentially filtered with a comb filter to reduce or remove the harmonics of the signal. However, the quality of the speech signal is badly affected.

In WO 2006/099670 A1 a device for monitoring the airway is described, wherein sound in the audible frequency range is introduced into the respiratory tract of an object and from the reflected or processed sound, the state of the respiratory tract is determined. For example, it is possible to detect airway obstruction. In a variant of the invention, the exceeding of specific threshold values is checked by means of the FFT (fast Fourier transformation), from which conclusions are drawn regarding the treatment of the measured signal.

It is an object of the invention to overcome the above-mentioned disadvantages of the prior art and to improve the speech quality of EL users using electronic mediators such as microphones.

1. a) Aufteilen eines einkanaligen Sprachsignals S(w,t), das aus der Summe der Komponenten eines zeitvarianten anteils x(w)H(w,t) und eines zeitinvarianten Anteils x(w)F(w) besteht, in eine Reihe von Frequenzkanälen durch Überführen vom Zeitbereich in einen diskreten Frequenzbereich,
2. b) Herausfiltern des zeitinvarianten Anteils x(w)F(w) durch Herausfiltern der Modulationsfrequenz 0 Hz mittels eines Hochpass- bzw. Notchfilters in jedem Frequenzkanal, und
3. c) Rücktransformieren des gefilterten Sprachsignals vom Frequenzbereich in den Zeitbereich und Zusammenführen zu einem einkanaligen Ausgangssignal.

This object is achieved by a method of the type mentioned according to the invention by the following steps:

1. a) dividing a single-channel speech signal S (w, t), which consists of the sum of the components of a time-variant component x (w) H (w, t) and a time-invariant component x (w) F (w), in a series of Frequency channels by transitioning from the time domain to a discrete frequency domain,
2. b) filtering out the time invariant component x (w) F (w) by filtering out the modulation frequency 0 Hz by means of a high pass filter in each frequency channel, and
3. c) inverse transforming the filtered speech signal from the frequency domain to the time domain and merging into a single channel output signal.

The invention makes use of an improved model of the use of an EL, according to which the EL fundamental sound articulated to a speech signal as well as the unaltered portions of the EL interfering with the perception of the speech signal come from a common source, namely the EL. Since the disturbing inarticulate fundamental noise of the EL in the modulation range is recognizable as a time-invariant signal, it can be easily filtered out by a suitable procedure. Thus, there is a separation not of signal sources, but of propagation paths (a propagation path through the articulation organs of a speaker, another propagation path from the application site on the neck of the speaker directly to the ear of the listener, or to the microphone or recording means).

The person skilled in the art knows a large number of possibilities for converting a digitized, single-channel signal into the frequency domain and thus dividing it into a series of frequency channels. In each frequency channel the modulation frequency of the EL is filtered by suitable filters - e.g. Notch or high pass filter applied to the amount - suppresses and thus improves the quality of the articulated signal components.

Similar prior art methods consider the articulated components as well as the unmodified components as coming from different sources, and adopt similar approaches to this model, for example filtering by means of bandpass filters which, however, also attenuate the speech signal.

The method according to the invention thus aims to increase the intelligibility of the language of EL users or to make the signal more pleasant and "more human". The aim is to reduce or eliminate the direct sound from the EL when communicating via electronic means (e.g., telephone).

The implementation of the method according to the invention can be done for example by a software plug-in, as a hard-wired solution or as an analog circuit.

From the large number of known methods for transferring a signal into the frequency range or back, the conversion in step a) of the method according to the invention is advantageously carried out by Fourier transformation and the inverse transformation in step c) by means of inverse Fourier transformation. The transfer takes place in blocks (eg blocks of 20 ms) at short intervals (refresh every 10 ms, for example). The division of the signal into a series of frequency channels takes place when transferring the signal into the frequency domain.

In a variant of the invention, the transfer of the speech signal in step a) and the inverse transformation in step c) with a corresponding filter bank.

The results of the method according to the invention can be further improved if signal compression takes place before the filtering in step b) and decompression takes place after step b). The compression can be prevented that at high amplitudes whose changes are so dominant that the changes of small amplitudes are not taken into account. Compression makes relative changes more visible to the filter.

In a further embodiment of the method according to the invention, before the inverse transformation in step c), a rectification of the negative signal components takes place.

Fig. 1

eine vereinfachte Darstellung der Verwendung eines EL und die auftretenden Signalpfade,

Fig. 2

eine vereinfachte Darstellung der Situation, in der die erfindungsgemäße Methode Anwendung findet und

Fig. 3

ein Blockschaltbild der erfindungsgemäßen Methode.

In the following the invention will be explained in more detail with reference to a non-limiting embodiment, which is illustrated in the drawing. In this shows schematically:

Fig. 1

a simplified representation of the use of an EL and the occurring signal paths,

Fig. 2

a simplified representation of the situation in which the method according to the invention finds application and

Fig. 3

a block diagram of the method according to the invention.

In Fig. 1 the different transmission paths of the signal of an EL 1 are sketched. In this case, an EL 1 is arranged on the neck of a speaker 2. The sound produced by the EL 1 spreads on the one hand through the normal speech channels (mouth and nose) 5 of the first speaker 2 and is there articulated to language; this first signal 3 is clearly variable, or time-variant. At the ear of a listener 4 comes next to this time-variant signal 3, a second signal 6 (in Fig. 1 dash-dotted lines) in the form of the direct sound of EL 1, this signal 4 is largely stationary and is therefore assumed to be invariant in time. The second part 6 of the total signal, that is, the background noise of the EL 1, is perceived by the listener 4 as an interference signal and reduces the intelligibility of the speech of the speaker 2. The original excitation by means of the EL 1 is thus transmitted via two different paths.

Although the invention relates to the improvement of the speech quality of an EL speaker when using electronic mediators - so instead of a listener, the signals would be recorded, for example, with a microphone. To illustrate the starting position However, for the sake of clarity, this general model has been chosen.

Fig. 2 shows a simplified model representation of the situation to which the inventive method for the suppression of a disturbing second signal 6 (see Fig. 1 ) is applied. It is readily apparent that in the method according to the invention there is no separation of signal sources, but of propagation paths.

A source signal x (w) from a signal source 7 propagates over two different signal paths. In the first signal path, the output signal is modulated by a time-variant filter H (w, t) to a time-variant signal x (w) H (w, t). In the second signal path, the output signal is changed only by a time-invariant filter F (w) to a signal x (w) F (w).

The signals of the two paths are then in a receiver 8 - eg the ear of a listener, a microphone or similar. - Summed to a signal available for measurement S (w, t). The signal then consists of the sum of the components, $$\mathrm{S}\left(\mathrm{w},\mathrm{t}\right)=\mathrm{x}\left(\mathrm{w}\right)\mathrm{H}\left(\mathrm{w},\mathrm{t}\right)+\mathrm{x}\left(\mathrm{w}\right)\mathrm{F}\left(\mathrm{w}\right)\mathrm{,}$$

It is now possible to separate the signal components from the time-invariant and time-variant signal paths, by attenuating either all signal components which change over time or are constant over time. For example, as a result, only the time-variant component S1 (w, t) ~ x (w) H (w, t) is obtained.

In the application for speech with EL, the inarticulate signal component x (w) F (w) (ie the background noise of the EL) superimposes the time-variant speech signal x (w) H (w, t) and thereby causes a loss of intelligibility for the speech signal. Speech intelligibility is improved by separating the time-variant signal component from the time-invariant signal component.

Fig. 3 shows a possible implementation of the method according to the invention. In this case, an arbitrary digital speech signal 9 from a speaker with EL can be present at the input. In a first step 10, using the short-time Fourier transformation, the speech signal 9 is transformed in blocks into the frequency domain and thus divided into a series of frequency channels. The person skilled in the art can choose here from various established methods for the transformation of a signal from the time domain into the frequency domain; In addition to the Fourier transformation, for example, the discrete cosine transformation is also used - however, the prerequisite for an application according to the invention is that the transformation is reversible. The signal is divided at a certain refresh rate (eg 10 ms) into blocks of, for example, 20 ms in length, which are each fanned out into a series of frequency channels 11. The originally single-channel speech signal 9 is thus split into a plurality of frequency ranges, which change as a result of time. The frequency signal is complex, but subsequently only the absolute value is modified, phase 15 remains unchanged.

In step 10, a filter bank may also be used, reducing the sample rate of the signal after the filter bank. The reduction of the sampling rate corresponds to the block formation when using the Fourier transformation.

In a further functional block 12, each frequency channel 11 is filtered, for example with a high-pass filter or notch filter. This filtering allows the filtering of certain frequencies - in the sound technique narrowband interference is eliminated with notch filters. Since the EL oscillates at a certain frequency - for example 100 Hz - the interference signal, which is not changed by the articulation organs of a speaker, results in the frequency range amplitudes in the 100 Hz channel with the modulation frequency 0 Hz - ie the amplitude of the EL Signal does not change. The interference signal is characterized in that it is perfectly time-invariant. To filter the background noise of the EL, a notch or a high pass filter are used. The limiting frequency for the high-pass filter is the modulation frequency of the EL; the notch filter is chosen so that it locks exactly at the modulation frequency of the EL.

In the real implementation of course a perfect temporal invariance due to reflections, refractions, ambient noise and structural needs of the EL will not be achievable. However, since the filter is not restricted to just one frequency but covers a specific frequency range-in this case a modulation frequency range-the function of the method according to the invention is ensured.

In a final function block 13, the feedback of the signals into the time domain, for example by means of inverse Fourier transformation and the merger of the frequency channels 11 back into a channel by means of overlap-add. The overlap-add method is a method of digital signal processing known to the person skilled in the art. The result is a single-channel output signal 14, in which the interference signal of the EL is filtered out or at least attenuated. The output signal can then be processed further.

When using a filter bank in step 10, the sampling rate of the signal is increased again after the filtering in step 12 and then treated further as described.

Basically, these statements represent only the most important components of the method according to the invention; before the filtering in block 12, the signal can be compressed, after the decompression can be provided. A rectification before the inverse transformation into the time domain can also be favorable, since unauthorized negative values can arise during processing.

The invention can be used, for example, as an additional device for telephoning. In a conventional analogue telephone, the device is simply integrated into the handset. In a telephone with integrated digital signal processor, the integration of the invention by a software plug-in is possible. Also, the realization in the context of a hardwired solution, e.g. also in an analog circuit, is possible.

The method according to the invention can also be used when using an EL in which two or more frequencies can be switched back and forth in order to give the speech a more realistic sound. This applies both to discrete frequency jumps and to continuous changes in the fundamental frequency assuming that the frequencies being switched are within a frequency band into which the fundamental signal is split.

The width of the modulation frequency filter determines how fast the frequency may change. With very slow, continuous changes, the frequency can change over the entire band of the frequency band if the suppression function works - the decisive factor is not the size but the speed of the change. When switching the EL on and off, which corresponds to rapid changes, the suppression only takes effect after a few milliseconds - depending on how broadly the notch filter is selected or where the fundamental frequency of the high-pass filter lies.

However, the changes in the fundamental frequency must not be too large. In order to ensure the function according to the invention, for example, the frequency channels into which the signal is split would have to be expanded, or the filtering by means of a high-pass filter would have to start at a somewhat higher frequency.

Claims (5)

1. Method for improving the speech quality of an electric larynx (EL) speaker, whose speech signal S(w,t) is digitised by suitable means, characterised by the following steps:

   a) dividing a single-channel speech signal S(w,t), which consists of the sum of the compounds of a time variant signal x(w)H(w,t) and a time-invariant signal x(w)F(w), into a series of frequency channels by transforming it from a time domain into a discrete frequency domain,

   b) filtering out the time-invariant signal x(w)F(w) by filtering out the modulation frequency 0 Hz by means of a high-pass or notch filter in each frequency channel, and

   c) back-transforming the filtered speech signal from the frequency domain into the time domain and combining it into a single-channel output signal.
2. Method according to claim 1, characterised in that the transforming of the speech signal in step a) is carried out by means of a Fourier transformation and the back-transformation in step c) is carried out by means of an inverse Fourier transformation.
3. Method according to claim 1, characterised in that the transformation of the speech signal in step a) and the synthesis of the frequency channels in step c) is carried out with a corresponding filter bank.
4. Method according to one of claims 1 to 3, characterised in that before the filtering in step b) a signal compression is carried out, and after step b) a decompression is carried out.
5. Method according to one of claims 1 to 4, characterised in that before the back-transformation in step c) a rectification of the negative signal components is carried out.

Images