JP2010026323A - Speech speed detection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech speed detection device which is hardly affected by environmental noise and can accurately detect the speech speed of a speaker. <P>SOLUTION: The speech speed detection device includes: a linear prediction coefficient variation calculation section 11 for calculating variation of a linear prediction coefficient by performing linear prediction on a speech signal; a first envelope calculation section 12 for calculating the envelope of a variation sum of the linear prediction coefficient; a first peak detection section 13 for detecting the peak of the envelope; a first speech speed calculation section 14 for calculating the number of peaks per unit time, from the number of detected peaks; a speech absolute value calculation section 21 for calculating the absolute value of the speech signal; a second envelope calculation section 22 for calculating the envelope; a second peak detection section 23 for detecting the peak of the envelope; a second speech speed calculation section 24 for calculating the number of peaks per unit time, from the number of detected peaks; and a comprehensive speech speed calculation section 10 for calculating the speech speed on the basis of both of the number of peaks per unit time. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a speech speed detection device that is used in a real-time communication device such as an interphone and detects the speaking speed (speaking speed) of a speaker for speech speed conversion.

  Conventionally, in a field such as an IC recorder, a digitized audio signal is compressed / expanded on the time axis, the compressed / expanded audio signal is converted into an analog signal and output from a speaker. Converting speed has been done. As is well known, the speaking speed varies depending on the speaker, and the speaking speed during that time is not constant even when the same speaker is speaking. Therefore, if speech speed conversion is performed on a speech signal at a constant compression / decompression rate, the reproduced speech may be faster or slower than the speech speed desired by the user (listener), which may be difficult for the user to hear. is there.

  For this reason, a method has been proposed in which the speaking speed of an actual speaker is detected and speaking speed conversion is performed by setting a compression / decompression ratio according to the detected speaking speed. For example, in Patent Document 1, smoothing processing is performed on an envelope of a time axis waveform of speech, the number of waveform peaks per unit time is counted to calculate speech speed, and speech speed conversion is performed according to the value. . However, since the change of the voice appears not only in the amplitude but also in the frequency, in such a method, the accuracy of detecting the speech speed is low, and the reproduced voice may be heard unnaturally.

By the way, in the case of a device that makes a call in real time, such as an intercom, the two parties who are making calls are often unfamiliar with each other, and it is often impossible to predict at what speed the other party will speak. , Likely to miss the content of the other person's story. For this reason, it is required to detect the speaking speed of the speaker more accurately in a real-time communication device.
Japanese Patent Laid-Open No. 7-64597

  The present invention has been made to solve the above-described problems of the conventional example, and an object of the present invention is to provide a speech speed detection device capable of more accurately detecting a speaker's speech speed.

  In order to achieve the above object, the invention according to claim 1 is a speech speed detection device, which performs linear prediction analysis on an input speech signal and obtains an envelope of the total amount of change of the obtained prediction coefficient. The peak number per unit time in the envelope is the first speech speed parameter, the envelope of the absolute value of the speech signal is taken, the peak number per unit time in the envelope is the second speech speed parameter, and the first speech speed parameter The speech speed is calculated based on the second speech speed parameter multiplied by a predetermined contribution rate and added.

  The invention according to claim 2 is a speech speed calculation device, which performs linear prediction on an input speech signal and calculates a change amount of a linear prediction coefficient, and the linear prediction count. A first envelope calculation unit for obtaining an envelope of a sum of changes in the linear prediction count obtained by a change amount calculation unit; a first peak detection unit for detecting a peak of the envelope obtained by the first envelope calculation unit; Calculating the number of peaks per unit time from the number of peaks detected by the first peak detector, and outputting the obtained number of peaks as a first speech speed parameter; and the input A sound absolute value calculating unit for obtaining an absolute value of the sound signal to be generated; a second envelope calculating unit for obtaining an envelope of the sound absolute value obtained by the sound absolute value calculating unit; The number of peaks obtained by calculating the number of peaks per unit time from the number of peaks detected by the second peak detection unit that detects the peak of the envelope obtained by the rope calculation unit and the second peak detection unit Is calculated as a second speech speed calculation unit, and the first speech speed parameter is calculated based on the first speech speed parameter and the second speech speed parameter multiplied by a predetermined contribution rate. And a comprehensive speech speed calculation unit.

  The invention according to claim 3 is the speech speed calculation device according to claim 2, further comprising a voice section detector for detecting a voice section from the input voice signal, and determining that the voice section detector is a voice section. The first speech speed calculation unit and the second speech speed calculation unit each calculate the number of peaks per unit time only for the section that has been performed.

  According to a fourth aspect of the present invention, in the speech speed calculation device according to the second aspect, the first peak detection unit and the second peak detection unit have a predetermined amount of change from a maximum value to a minimum value of an envelope. If the value is smaller than the maximum value, the maximum value is not detected as a peak.

  According to the first or second aspect of the present invention, it is possible to extract changes such as the frequency characteristics of speech from the envelope of the change amount of the linear prediction coefficient. On the other hand, the change amount of the linear prediction coefficient varies little with respect to a signal having a small amplitude, and there may be a case where a change in voice feature cannot be extracted. For this reason, it is possible to realize high-accuracy speech speed detection by combining an envelope of a speech absolute value that can extract a change in speech with a low signal level.

  According to the invention of claim 3, since the speech speed detection processing is performed only for the speech section in which the speaker is actually speaking in the speech signal, it is caused by environmental noise in the non-speech section where the speaker is not speaking. The influence can be eliminated and more accurate speech speed detection can be performed.

  According to the invention of claim 4, since only a peak having a large change is detected, the influence of noise or the like can be reduced, and speech speed detection with high accuracy can be performed.

  A speech speed detection apparatus according to an embodiment of the present invention will be described with reference to the drawings. First, an example of use of the speech speed detection device 1 is shown in FIG. For example, an audio signal input from a microphone (not shown) or the like is input to the speech speed detection device 1 and the processing described below is performed to detect the speaker's speech speed. The detected speech speed is input to the speech expansion rate determination unit 2 as a speech speed parameter, and the speech expansion rate is determined. The speech signal is input to the speech speed conversion unit 3 in parallel with the speech speed conversion device 1, and the speech speed of the speech signal is converted based on the speech expansion rate determined by the speech expansion rate determination unit 2. Then, the voice signal that has been subjected to the speech speed conversion is converted into voice from the speaker 4 and output. The speech speed conversion device 1 may be configured as part of a circuit of a communication device having a speech speed conversion function such as a telephone or an interphone, or may be modularized as a dedicated IC.

  FIG. 2 shows a configuration example of the speech speed detection apparatus 1 according to the present embodiment. FIG. 3 shows another configuration example of the speech speed detection device 1. The speech speed detection device 1 is obtained by a linear prediction count change amount calculation unit 11 that performs linear prediction on an input speech signal and calculates a change amount of a linear prediction coefficient, and a linear prediction count change amount calculation unit 11. A first envelope calculation unit 12 for obtaining an envelope of the total amount of change of the linear prediction count, a first peak detection unit 13 for detecting a peak of the envelope obtained by the first envelope calculation unit 12, and a first peak detection unit 13 calculates the number of peaks per unit time from the number of peaks detected by 13 and outputs the obtained number of peaks as a first speech speed parameter; and the absolute value of the input voice signal A sound absolute value calculation unit 21 for obtaining a value, a second envelope calculation unit 22 for obtaining an envelope of a sound absolute value obtained by the sound absolute value calculation unit 21, and a second envelope performance The number of peaks obtained by calculating the number of peaks per unit time from the number of peaks detected by the second peak detection unit 23 and the second peak detection unit 23 that detect the peak of the envelope obtained by the unit 22 Is calculated as a second speech speed parameter, and the first speech speed parameter is calculated based on the first speech speed parameter and the second speech speed parameter multiplied by a predetermined contribution rate. The integrated speech speed calculation unit 10 is provided. In the configuration example illustrated in FIG. 2, a voice section detection unit 20 that further detects a voice section from an input voice signal is provided.

  These linear prediction count change amount calculation unit 11, first envelope calculation unit 12, first peak detection unit 13, first speech speed calculation unit 14, speech absolute value calculation unit 21, second envelope calculation unit 22, second peak detection The unit 23, the second speech speed calculation unit 24, the total speech speed calculation unit 10, and the voice section detection unit 20 may be configured by individual circuits, or may be configured by the same CPU, ROM, RAM, and the like. Alternatively, predetermined processing is performed on the digitized audio signal.

  As is clear from FIG. 2 or FIG. 3, the speech speed detection apparatus 1 according to the present embodiment performs linear prediction analysis on the input speech signal, and obtains the envelope of the total amount of change of the obtained prediction coefficient. The first speech speed calculation system that outputs the peak number per unit time in the envelope as the first speech speed parameter and the absolute value envelope of the audio signal are taken, and the peak number per unit time in the envelope is the second speech speed. The second speech speed calculation system that outputs as a parameter is provided, and finally the speech speed is calculated based on the first speech speed parameter and the second speech speed parameter multiplied by a predetermined contribution rate and added. Calculate.

The linear prediction count change amount calculation unit 11 inputs an audio signal to the FIR filter and obtains a linear prediction coefficient by using an LMS algorithm or the like for the filter coefficient. The total sum he (n) of the temporal change amount of the linear prediction coefficient in the M-th order linear prediction filter at time n is obtained from the following equation. h _m (n) is the m-th linear prediction coefficient at time n.

  The first envelope calculation unit 12 takes an envelope of the total amount of change in the prediction coefficient, the first peak detection unit 13 detects the number of envelope peaks as a syllable feature amount, and the first speech speed calculation unit 14 The number of peaks per unit time is counted, and the counted number of peaks per unit time is output as the first speech speed parameter. When linear prediction analysis is performed on a stationary signal, the obtained linear prediction coefficient does not change with time and becomes a constant value. Therefore, even if the input signal is a speech signal superimposed with stationary noise, Since the feature amount can be extracted, the speech speed can be stably detected even in a noisy environment.

  On the other hand, the audio absolute value calculation unit 21 calculates the absolute value of the input audio signal, and the second envelope calculation unit 22 takes the envelope of the absolute value of the audio signal. The second peak detection unit 23 detects the number of envelope peaks as a syllable feature amount, and the second speech speed calculation unit 24 counts the number of peaks per unit time, and the counted number of peaks per unit time is the second story. Output as a speed parameter. The total speech speed calculation unit 10 gives predetermined contribution rates (for example, 3: 7, 4: 6, 5: 5, 6: 4, 7: 3, etc.) to the first speech speed parameter and the second speech speed parameter, respectively. The speech speed is calculated based on the result of multiplication and addition.

  FIG. 4A shows an example of the time waveform of the input signal, and FIG. 4B shows an example of the waveform obtained by applying an envelope to the sum of the time waveform of the input signal and the time variation of the linear prediction coefficient. An example of a waveform obtained by applying an envelope to the signal absolute value of the time waveform of the input signal is shown in c).

  As described above, it is possible to extract changes such as the frequency characteristics of speech from the envelope of the change amount of the linear prediction coefficient. However, the change amount of the linear prediction coefficient varies little for a signal having a small amplitude, It may not be possible to extract changes in the characteristics of For example, focusing on the circle A in FIG. 4B and the circle B in FIG. 4C, when the envelope of the change amount of the linear prediction coefficient is used for detecting the speech speed, it cannot be extracted from the envelope of the signal absolute value. Such a change in voice can be detected. On the other hand, when attention is paid to the circle C in FIG. 4A, the circle D in FIG. 4B, and the circle E in FIG. 4C, the linear prediction coefficient of the portion where the speech amplitude is small (circle C) is shown. A change cannot be extracted from the change amount envelope (circle D), but a change in the sound can be extracted from the signal absolute value envelope (circle E). In this way, by using the envelope of the change amount of the linear prediction coefficient and the envelope of the signal absolute value in combination, the other drawback is complemented with each other, so that it is possible to realize highly accurate speech speed detection.

  In the configuration example shown in FIG. 3, the voice section is detected from the input voice signal by the voice section detection unit 20, and only for the section determined by the voice section detection unit 20 to be a voice section, that is, the voice section. Only during the period when the flag is on, the first speech speed calculator 14 and the second speech speed calculator 24 each calculate the number of peaks per unit time. In other words, the speech speed detection process is performed only for the speech section of the speech signal that the speaker is actually speaking, so it is possible to eliminate the influence of environmental noise in the non-speech section where the speaker is not speaking. Therefore, more accurate speech speed detection can be performed. Note that the speech segment detection unit 20 performs distinction between speech segments and non-speech segments and detection of speech segments by the method described in Japanese Patent Application Laid-Open No. 2005-156886, and detailed description thereof is omitted.

  As shown in FIG. 5, in the first peak detection unit 13 and the second peak detection unit 23, when the amount of change from the maximum value of the envelope to the minimum value is smaller than a predetermined set value (threshold value), the maximum You may comprise so that a value may not be detected as a peak. In the envelope of the change amount of the linear prediction coefficient and the envelope of the signal absolute value, the maximum value and the minimum value continue alternately. Focusing on the peaks marked with x in FIG. 5, even if the change from the minimum value to the maximum value is relatively large, the change from the maximum value to the minimum value is very small. Therefore, if the amount of change from the maximum value to the minimum value is compared with a predetermined set value, and if the change amount is smaller than the set value, it is not counted as a peak, only the peak with a large change is detected and noise etc. As a result, the speech speed can be detected with high accuracy.

  The speech expansion rate determination unit 2 shown in FIG. 1 determines a speech expansion rate when reproducing a speech signal based on the speech speed parameter calculated by the general speech speed calculation unit 10. Further, the speech speed conversion unit 3 performs speech speed conversion of the speech signal based on the speech expansion rate determined by the speech expansion rate determination unit 2. As the speech speed calculation algorithm, for example, a PICOLA (Pointer Interval Controlled OverLap and Add) algorithm can be used.

  In addition, this invention is not limited to description of the said embodiment, A various deformation | transformation and application are possible in the range which does not deviate from the meaning of invention. For example, it goes without saying that the present invention can be used for detecting the speech speed of not only a real-time communication device such as a telephone or an interphone but also an audio playback device such as an IC recorder.

The block diagram which shows the usage example of the speech-speed detection apparatus which concerns on one Embodiment of this invention. The block diagram which shows the example of 1 structure of the speech-speed detection apparatus which concerns on this embodiment. The block diagram which shows the other structural example of the speech speed detection apparatus which concerns on this embodiment. (A) is a figure which shows the time waveform of an input signal, (b) is a figure which shows an example of the waveform which applied the envelope to the time waveform of the input signal, (c) is a time change of the time waveform of an input signal, and a linear prediction coefficient. The figure which shows an example of the waveform which applied the envelope with respect to the sum total of quantity. The wave form diagram for demonstrating the modification which was not counted as a peak, when the change from the maximum value to the minimum value in an envelope was small.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Speech speed detection apparatus 10 Comprehensive speech speed calculation part 11 Linear prediction coefficient variation | change_quantity calculation part 12 1st envelope calculation part 13 1st peak detection part 14 1st speech speed calculation part 20 Voice area detection part 21 Voice absolute value calculation part 22 Second envelope calculator 23 Second peak detector 24 Second speech speed calculator

Claims (4)

Perform linear prediction analysis on the input speech signal, find the envelope of the total amount of change in the obtained prediction coefficient, use the number of peaks per unit time in the envelope as the first speech speed parameter,
Taking the envelope of the absolute value of the audio signal, the peak number per unit time in the envelope as the second speech speed parameter,
2. A speech speed detecting apparatus, comprising: calculating a speech speed based on a result obtained by multiplying the first speech speed parameter and the second speech speed parameter by a predetermined contribution rate, respectively. A linear prediction count change amount calculation unit that performs linear prediction on an input speech signal and calculates a change amount of a linear prediction coefficient;
A first envelope calculation unit for obtaining an envelope of the sum of the change amounts of the linear prediction count obtained by the linear prediction count change amount calculation unit;
A first peak detector that detects a peak of the envelope obtained by the first envelope calculator;
A first speech speed calculator that calculates the number of peaks per unit time from the number of peaks detected by the first peak detector, and outputs the obtained peak number as a first speech speed parameter;
An audio absolute value calculation unit for obtaining an absolute value of the input audio signal;
A second envelope calculation unit for obtaining an envelope of the voice absolute value obtained by the voice absolute value calculation unit;
A second peak detector for detecting the peak of the envelope obtained by the second envelope calculator;
A second speech speed calculator that calculates the number of peaks per unit time from the number of peaks detected by the second peak detector, and outputs the obtained number of peaks as a second speech speed parameter;
A speech speed calculation comprising an overall speech speed calculation unit for calculating a speech speed based on a result obtained by multiplying the first speech speed parameter and the second speech speed parameter by a predetermined contribution rate, respectively. apparatus.   A voice section detector for detecting a voice section from the input voice signal is further provided, and the first speech speed calculator and the second episode are only applied to a section determined by the voice section detector as a voice section. The speech speed calculation device according to claim 2, wherein each of the speed calculation units calculates the number of peaks per unit time.   The first peak detection unit and the second peak detection unit do not detect the maximum value as a peak when the amount of change from the maximum value of the envelope to the minimum value is smaller than a predetermined set value. The speech speed calculation device according to claim 2.

Images