JPH09152889A - Speech speed transformer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a speech speed transformer capable of obtain the output voice having low deterioration by respectively providing a specific speech speed converting means a specific degree of similarity calculating means and a specific control means. SOLUTION: A section judging part 42 classifies an input voice based on the degree of similarity of waveforms being in the vicinities of the input voice signal to transmit the classified results to a signal processing part 43. The input voice signal is constituted of voice sections and non-voice sections and there are vowel section, consonant sections and transition sections from vowels to consonants or from consonants to vowels in the voice section. Then, when the section of the input voice signal is judged to be the transition section by the section judging part 42, the voice signal is compressed with a reference compression rate α determined by the multiplication of a reproducing speed by a time base compressing and extending part 51. Moreover, the input device signal of the vowel section and the input voice signal of the consonant section are respectively compressed with a compression rate αs smaller than the reference compression rate α and a compression rate αs larger than the α. That is, at the time of a high-speed reproducing of a VTR, etc., when the degree of similarity of waveforms calculated in these sections is large, the degree of compressions is made large and when the degree of similarity is small, the degree of compressions is made small.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice speed conversion device for converting the voice speed of a voice signal, and for example, a fast listening of a voice of a laser disc, a VTR, a TV, a TV telephone, a digital video disc or the like accompanied by an image. Audio playback device for slow listening, radio with hearing aids that converts audio signals into slow and easy-to-hear voices for hearing impaired people, telephones, hearing aids, English voices spoken at native speed into slow and easy-to-hear The present invention relates to an English learning device for conversion, a tape recorder for performing early or late listening of voice, a stereo system, a CD player, an MD player, a voice speed conversion device used in a voice guidance system and the like.

[0002]

2. Description of the Related Art A technique for converting the speech speed of an audio signal by subjecting an input audio signal to a time axis compression / expansion process has already been developed. The input voice is divided into a voice section and a non-voice section. The voice section is divided into a vowel section, a consonant section, and a transition section between the vowel section and the consonant section.
The voice signal in the vowel section has periodicity, and even if the time axis compression / decompression process is performed, the distortion during processing is small and the reproduced voice is hardly deteriorated. However, the voice signal in the consonant section has a waveform close to white noise, and when the time axis compression / expansion process is performed, the distortion during processing is large and the reproduced voice is deteriorated.

[0003]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech speed conversion device which can obtain output voice with little deterioration.

[0004]

A speech speed conversion apparatus according to the present invention comprises a speech speed conversion means including a time axis compression / expansion means for time axis compression / expansion processing of an input speech signal, and a waveform in the vicinity of the input speech signal. It is characterized in that it is provided with a similarity calculation means for calculating the similarity and a control means for controlling the speech speed conversion means based on the calculated similarity. A determination unit that determines whether or not the input voice signal is a voice signal in a silent section, and a unit that deletes a voice signal in a silent section whose duration is equal to or longer than a predetermined length may be provided.

As the control means, for example, a means for controlling the speech speed conversion means is used so that the degree of compression or expansion by the time axis compression / expansion means becomes smaller as the input voice signal having a smaller degree of waveform similarity. To be

As the control means, for example, the speech speed conversion means is controlled so that the degree of compression or the degree of expansion by the time-base compression / expansion means increases as the input voice signal having a higher degree of waveform similarity. Is used.

As the control means, for example, based on the calculated similarity, the input voice signal
The consonant section, the transition section between the vowel section and the consonant section, and the noise section are classified, and the input speech of the vowel section is compressed or expanded with a large degree of compression or expansion so that the consonant section can be compressed. The input audio in the transition section is compressed and expanded so that the input audio is compressed or expanded with a small compression or expansion degree. What controls the speech speed conversion means so as to perform processing so as to delete the input voice in the noise section is used.

As the control means, for example, based on the calculated similarity, the input voice signal
If the similarity is large in the consonant section, the transition section between the vowel section and the consonant section, and within the classified vowel section, the consonant section, or the transition section between the vowel section and the consonant section. In order to increase the degree of compression or expansion of the section or to reduce the degree of compression or expansion of the section when the degree of similarity is small, a unit that controls the speech speed conversion means is used.

It is considered that the smaller the similarity of the waveform in the vicinity of the input voice is, the closer the waveform of the input voice is to a white noise without periodicity. Therefore, by reducing the degree of compression or decompression of the time axis compression / decompression processing unit as the degree of similarity in the vicinity is smaller, it is possible to prevent deterioration of the reproduced voice.

[0010]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows the structure of a speech speed converter.

The speech speed conversion device comprises a voice signal input unit 41, a section discrimination unit 42, a signal processing unit 43, a voice memory 44 and a voice signal output unit 46. The signal processing unit 43
A time axis compression / expansion unit 51, a deletion unit 52 and the like are provided. The section determination unit 42 includes a waveform similarity calculation unit 61, a duration calculation unit 62, and a section classification unit 63.

The audio signal input section 41 includes, for example, an amplification section, an A / D conversion section, a frame memory and the like. The signal input to the audio signal input unit 41 is amplified,
It is converted into a digital signal and stored in the frame memory. The output of the audio signal input unit 41 is sent to the section discrimination unit 42 and the signal processing unit 43. In this embodiment, the case where an analog signal is input to the speech speed conversion device is shown.
A digital signal read from an IC memory or the like may be input to the speech speed conversion device. In this case, it is not necessary to provide the audio signal input unit 41 with an A / D conversion unit.

The section discriminating section 42 classifies the input speech based on the similarity of the waveforms in the vicinity of the input speech signal, and sends the classification result to the signal processing section 43. The input voice signal includes a voice section and a non-voice section. The voice section includes a vowel section, a consonant section, and a vowel-to-consonant or consonant-to-vowel transition section. The non-voice section includes a noise section composed of a stationary noise section and a non-stationary noise section, and a silent section.

In the vowel section, since the voice signal waveform has a periodic waveform, the similarity of nearby waveforms becomes large. In the consonant section, the voice signal waveform becomes a waveform close to white noise, so that the similarity of nearby waveforms becomes small. Also, in the transition period from the consonant to the vowel and the transition period from the vowel to the consonant, the similarity of the nearby waveform is intermediate between the similarity in the vowel section and the similarity in the consonant section.

In the stationary noise section, since the voice signal waveform has a periodicity, the similarity of nearby waveforms is large, but the duration is longer than in the vowel section. In the non-stationary noise section, the similarity of nearby waveforms is small, but the duration is longer than in the consonant section.

The waveform similarity calculator 61 first calculates the similarity of nearby waveforms and classifies the similarity into three levels of large, medium and small. The duration calculating unit 62 calculates the duration of the state of the determined degree of similarity. The section classification unit 63 includes a result of classification by the waveform similarity calculation unit 61 and a continuation length calculation unit 62.
The input speech signal is classified into a stationary noise section, a non-stationary noise section, a vowel section, a transition section, and a consonant section based on the duration calculated by the signal processing unit 43.
Send to

That is, when the degree of similarity is medium, it is determined that the input voice signal is in the transition section. If the degree of similarity is large and the duration of the state is within the predetermined time,
The input voice signal is determined to be in the vowel section. When the degree of similarity is high and the duration of the state is longer than the predetermined time, it is determined that the input audio signal is in the stationary noise section. If the degree of similarity is small and the duration of the state is within the predetermined time, it is determined that the input voice signal is in the consonant section. When the degree of similarity is small and the duration of the state is longer than the predetermined time, it is determined that the input voice signal is in the non-stationary noise section.

As a method of calculating the degree of similarity of waveforms in the vicinity,
The following method is used.

(1) Difference in amplitude between waveforms or difference of 2
Method of calculating the power In this case, the larger the difference amount, the smaller the degree of similarity.

(2) Method of calculating average value of amplitude difference between waveforms or square of average value In this case, the greater the average value of difference, the smaller the similarity.

(3) Method of calculating cross-correlation value between waveforms In this case, the greater the cross-correlation value, the greater the degree of similarity. The cross-correlation function R (τ) between the signal waveform x (t) and the signal waveform y (t) is calculated, for example, based on the following formula 1.

[0023]

(Equation 1)

(4) A method of calculating the difference value of the spectrum between the waveforms or the square of the difference value by frequency analysis. In this case, the larger the difference amount, the smaller the degree of similarity.

(5) A method of calculating the difference value or the square of the difference value between the waveforms of signal energy (power) by frequency analysis. In this case, the larger the difference amount, the smaller the degree of similarity.

The signal processing section 43 performs the following processing.

(A) First, a case will be described in which the speech speed conversion device is used in a reproduction device such as a VTR and the speech speed of a sound reproduced at high speed is reduced.

(1) The deleting unit 52 deletes the input speech signal which is judged to be a stationary noise section or a non-stationary noise section by the section judging unit 42.

(2) The input voice signal determined to be the transition section by the section determination unit 42 is the time-base compression / decompression unit 5
1, the compression processing is performed at the reference compression rate α determined by the reproduction speed magnification n. The reference compression rate αo is set to 1 / n or more, where n is the reproduction speed magnification. When the input audio signal is compressed at a compression rate of 1/2 during the VTR double speed reproduction, the two-pitch cycle is thinned out to one pitch cycle, and the output audio becomes twice the standard audio speed. However, the pitch remains unchanged.

When the reproduction speed is double speed, the reference compression rate αo is set to a value of 1/2 or more, for example, 2/3. That is, the 3-pitch cycle is thinned out to the 2-pitch cycle. In this case, the output voice speed is 3/2 times the standard voice speed. However, the pitch remains unchanged.

(3) The input voice signal determined by the section discriminating unit 42 to be in the vowel section is the time axis compression / decompression unit 5
1, the compression processing is performed at a compression rate αs smaller than the reference compression rate αo. That is, the degree of compression of the input voice determined to be the vowel section is made higher than that of the input voice determined to be the transition section. Therefore, the output voice speed for the input voice determined to be the vowel section is faster than the output voice for the input voice determined to be the transition section.

(4) The input voice signal determined by the section discriminating unit 42 to be a consonant section is the time axis compression / decompression unit 5
1, the compression processing is performed at the compression rate αs larger than the reference compression rate αo. That is, the degree of compression of the input voice determined to be the consonant section is made smaller than that of the input voice determined to be the transition section. Therefore, the output voice speed for the input voice determined to be the consonant section is slower than the output voice for the input voice determined to be the transition section.

(B) A case in which the speech speed conversion device is used in the hearing aid device and the speech speed of the voice reproduced at the standard speed is reduced will be described.

(1) The deleting unit 52 deletes the input speech signal which is judged to be the stationary noise section or the non-stationary noise section by the section judging unit 42.

(2) The input voice signal determined to be the transition section by the section determination unit 42 is the time axis compression / decompression unit 5
1, the extension processing is performed at the reference extension rate βo. The reference expansion rate βo is set to 3/2, for example. That is,
The 2-pitch cycle is extended to the 3-pitch cycle. In this case, the output voice speed is ⅔ times the standard voice speed.

(3) The input voice signal determined by the section discriminating unit 42 to be a vowel section is the time axis compression / decompression unit 5
1, the extension processing is performed at the extension rate β1 larger than the reference extension rate βo. β1 is set to 2/1, for example. That is, one pitch period is extended to two pitch periods,
The output voice speed in this case is 1/2 times the standard voice speed. That is, the degree of expansion of the input voice determined to be the vowel section is made higher than that of the input voice determined to be the transition section. Therefore, the output voice speed for the input voice determined to be the vowel section is slower than the output voice for the input voice determined to be the transition section.

(4) The time-axis compression / expansion unit 51 expands the input voice judged by the section judgment unit 42 to be a consonant section at an expansion rate β2 smaller than the reference expansion rate βo. In other words, the input voice determined to be the consonant section is
The degree of expansion is reduced as compared with the input voice determined to be the transition section. Therefore, the output voice speed for the input voice determined to be the consonant section is faster than the output voice for the input voice determined to be the transition section.

The example of the high speed reproduction of the VTR and the hearing aid has been described above. In the above-described embodiment, the degree of compression / expansion is constant in the classified vowel section, the consonant section, and the transition section between the vowel section and the consonant section. However, in the vowel section, the consonant section, or the transition section between the vowel section and the consonant section, the degree of compression / expansion may be changed based on the similarity of the waveform in the section.

That is, during high-speed reproduction of a VTR or the like, the degree of compression is increased if the degree of similarity of the waveform calculated in the section is large, and the degree of compression is decreased if the degree of similarity is small. In the hearing aid application, the degree of extension is increased if the degree of similarity calculated within the section is large, and the degree of extension is decreased if the degree of similarity is small.

As the time axis compression / expansion method used in the time axis compression / expansion unit 51, for example, an overlap addition method by pointer movement control (Pointer Interval Control Overlap and
Add: PICOLA), TDHS (Time Domain Harmonic Scaling)
There are laws etc.

A method of compressing an input signal (input audio data to the time axis compression / expansion unit 51) at a compression rate of 2/3 using PICOLA will be briefly described with reference to FIG.
First, the pitch period is extracted from the input signal. The extracted pitch period is Tp. A weight (weighting function K1) that linearly goes from 1 to 0 is added to the waveform A to create the waveform A ′. 0 to 1 for waveform B
A weight (weighting function K2) toward
Is created.

Then, these waveforms A'and B'are added together to create a waveform A '* B' of length Tp.
These weights are added to maintain continuity at the connection points before and after the waveform A ′ * B ′. Then the pointer becomes
The same operation is performed after moving by 3 Tp, which is the length determined based on the compression rate. As a result, two waveforms A ′ * B ′ and C are obtained from the three waveforms A, B, and C. In this manner, a signal for three pitch periods is compressed into a signal for two pitch periods.

The output of the signal processing unit 43 is the audio memory 44.
After being temporarily stored in, the audio signal is output to the audio signal output unit 46. The audio signal output unit 46 includes a D / A conversion unit. The digital signal sent from the audio memory 44 to the audio signal output unit 46 is converted into an analog signal and output from the audio signal output unit 46. In this example,
Although the case where the voice signal is output as an analog signal from the voice speed conversion device has been shown, the voice signal may be output as a digital signal from the voice speed conversion device. In this case, it is not necessary to provide the audio signal output unit 46 with the D / A conversion unit.

When both the input and output signals of the voice speed converter are analog signals, D in the voice signal output section 46 is used.
The sampling frequency of the A / A converter is set to the standard sampling frequency f _SO, and the A / D in the audio signal input unit 41 is set.
The sampling frequency of the converter is set to n · f _SO , where n is the reproduction speed multiplication factor. Therefore, even during high-speed reproduction, the pitch of the output voice becomes the original pitch.

When both the input and output signals of the speech speed converter are digital signals, assuming that the reproduction speed multiplication factor is n, the audio signal is output with respect to the output speed of the data output from the audio signal output unit 46. The input speed of the data input to the input unit 41 is set to be n times. Therefore, even during high-speed reproduction, the pitch of the output voice becomes the original pitch.

Note that it may be determined whether or not there is a silent section in the non-voice section. Then, the voice signal in the silent section whose duration is equal to or longer than the predetermined length may be deleted. This has the advantage that an increase in the amount of storage in the audio memory can be suppressed and the memory overflow can be reduced.

The judgment as to whether or not it is a silent section is made by whether or not the power average of a predetermined number of audio data from the audio signal input unit 41 is smaller than a given threshold value. That is, if the power average is smaller than the given threshold value, it is determined as a silent section.

More specifically, the average power value P of a predetermined number of audio data from the audio signal input unit 41 is calculated. This average power value P is the amplitude of each of a predetermined number of sampled audio data i ₀ , i ₁ , ... i _{N -1} (N
Is a predetermined number of audio data), and is calculated by the following formula 2.

[0049]

(Equation 2)

The silent section whose duration is less than the predetermined length is
After calculating the degree of similarity, the voice speed may be converted, or the voice speed may be converted at a certain compression / expansion rate.

[0051]

According to the present invention, output voice with little deterioration can be obtained.

[Brief description of the drawings]

FIG. 1 is a configuration diagram showing a schematic configuration of a speech speed conversion device.

FIG. 2 is a schematic diagram for explaining a time axis compression / expansion method using PICOLA.

[Explanation of symbols]

 41 voice signal input unit 42 section discrimination unit 43 signal processing unit 44 voice memory 46 voice signal output unit 51 time axis compression / expansion unit 52 deletion unit 61 waveform similarity calculation unit 62 continuation length calculation unit 63 section classification unit

Claims (6)

[Claims] 1. A speech speed conversion means including a time axis compression / expansion means for time axis compression / expansion processing of an input voice signal, a similarity degree calculation means for calculating a similarity degree of a waveform in the vicinity of the input voice signal, and a calculated rate. A speech speed conversion device, comprising: a control means for controlling the speech speed conversion means based on the similarity. 2. The control means controls the speech speed conversion means such that the degree of compression or the degree of expansion by the time axis compression / expansion means becomes smaller as the input voice signal having a smaller degree of waveform similarity. Item 1. The speech speed conversion device according to item 1. 3. The control means controls the speech speed conversion means such that the degree of compression or the degree of expansion by the time axis compression / expansion means becomes greater as the input voice signal having a higher degree of waveform similarity. Item 1. The speech speed conversion device according to item 1. 4. The control means classifies the input voice signal into a vowel section, a consonant section, a transition section between a vowel section and a consonant section, and a noise section based on the calculated similarity, and Transition such that the degree of compression or decompression is increased for input speech to perform compression / decompression processing, and the degree of compression or decompression for input speech in the consonant section is reduced to perform compression / decompression processing. The speech speed conversion means is designed to perform compression / expansion processing on the input speech in the interval with the degree of compression or expansion by the time axis compression / expansion means being moderate, and to delete the input speech in the noise interval. The speech speed conversion apparatus according to claim 1, which controls the speech speed. 5. The control means classifies the input voice signal into a vowel section, a consonant section, and a transition section between the vowel section and the consonant section based on the calculated similarity, and within the classified vowel section, In the consonant section or in the transition section between the vowel section and the consonant section, when the degree of similarity is large, the degree of compression or expansion of the section is large, and when the degree of similarity is small, the compression of the section is performed. Alternatively, the speech speed conversion means is controlled so that the degree of expansion becomes small.
The speech speed conversion device described in. 6. The method according to claim 1, further comprising: a determining unit that determines whether or not the input voice signal is a voice in a silent section, and a unit that deletes a voice signal in a silent section having a duration of a predetermined length or more. 6. The speech speed conversion device according to any one of 3, 4, and 5.