JPH0916193A - Speech speed converter - Google Patents

Abstract

(57) [Summary] [Purpose] The consonant part of speech is emphasized at the same time as speech speed conversion. A consonant detection means for detecting a consonant portion from a finer power pattern in the frame and a consonant emphasis processing means for amplifying the consonant portion in a frame for which waveform expansion processing is not performed by power determination processing for each frame. Are provided in the speech speed conversion device. [Effect] The consonant emphasis processing can be realized at the same time as the speech speed conversion processing.

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 a voice speed, and particularly, while performing a voice speed conversion process,
It relates to a method of emphasizing a consonant part of a voice.

[0002]

2. Description of the Related Art As a means for assisting the hearing of a hearing-impaired person, an analog type hearing aid which processes an amplitude and a frequency characteristic of a voice by using an analog circuit has been mainly used. On the other hand, in recent years, research and development for applying digital signal processing to compensation of hearing impairment have been actively conducted. Regarding this research and development trend, for example, the Journal of the Acoustical Society of Japan (1991
Vol. 47, No. 10, P760-P765) "Application of Digital Technology to Compensate for Hearing Impairments" and "Speech-perception aids for hearing-
impaired people: Current status and needed reserac
h ", J.of Acoust. Soc. America, Vol.90, No.2, P
t.1, Aug. 1991., etc.

Generally, in order to compensate for the loss of hearing, the sound pressure level is amplified and the dynamic range is compressed for each frequency according to the hearing characteristics of the user. In a conventional analog hearing aid, such processing is realized by an analog circuit. In addition, in digital hearing aids that have been developed in recent years,
By implementing this processing with software such as a digital filter, the user's hearing characteristics can be adapted in more detail.

In response to such a trend, in recent years, only the speed is changed by digital signal processing without changing the pitch of the voice, and hearing is performed over the entire auditory system including a decline in higher-order language processing speed. Attempts have been made to drop. For such a speech speed conversion technique, for example, "Assessment of speech speed conversion speech by a hearing-impaired person" by the Acoustical Society of Japan (March 1994, 2-4-7) and the Acoustical Society of Japan (50 Volume 7
No., P509-P520) "Real-time speech speed conversion type listening system" and the like.

In particular, recently, a portable voice speed conversion device has been developed in which this voice speed conversion function is contained in a portable device. For this device, refer to "A portable digital spee
ch-rate converter and its evaluation by hearing-im
paired listeners ", in Proc. of Int. Conf. on Spok
en Language Processing (ICSLP94), pp.2055-2058, Y
okohama, Sep. 1994.

FIG. 4 shows the flow of signal processing of this conventional speech speed conversion apparatus. The thick solid line in the figure indicates the flow of the audio signal,
The dotted line indicates the flow of control signals.

In the above-mentioned conventional speech speed conversion apparatus, the voice inputted through the microphone 11 is converted into the analog amplifier 1
2. After passing through the low-pass filter 13, the A / D converter 14
To a digital value. Digitized voice has a fixed time length (hereinafter referred to as frame) 2
The signals are alternately input to the two input frame buffers 21 and 22. While inputting to one of the input frame buffers, the voice data of the other input frame buffer is subjected to a voice speed conversion process 30 for expanding or contracting the time axis of the voice without changing its pitch. Then, the processing result is recorded in the output ring buffer 50 having a relatively large capacity. The output ring buffer 50 is a memory whose addresses are circularly used so that the end address is followed by the start address. Independently of writing the speech speed conversion processing result to the output ring buffer 50, the processed data is taken out from the output ring buffer 50 one by one, and D / A
Converter 64, low-pass filter 63, analog amplifier 6
By simultaneously converting and outputting the sound through the speaker 2 and the speaker or the earphone 61, the speech speed conversion processing and the output of the conversion result proceed at the same time, and the real-time speech speed conversion is realized. Therefore, the output voice is output with a delay time with respect to the input voice, and the delay time also spreads with time. That is, the listener can listen to the voice slowly, but slowly behind the reality.

This speech speed conversion operation is controlled by the user through the slow switch 41 provided on the apparatus, the speech speed conversion is performed while the slow switch is pressed, and the speech speed is converted while the slow switch is not pressed. The speed conversion is not applied, and the input voice is output as it is. Switch means 40 controlled by slow button 41 in FIG.
This behavior is represented by.

On the other hand, the speech speed conversion processing 30 is composed of a power judgment processing 31 and a waveform expansion processing 32. The waveform expansion processing 32 is applied only to the waveform of the frame having a large power selected by the power judgment processing 31, and is selected. The part that is not processed operates so that it is transferred to the output buffer as it is.

FIG. 5 shows the procedure of this speech speed conversion processing. In the power judgment processing 31, the threshold value Th is compared with the power of the audio signal of each frame. As a result, the waveform expansion processing 32 is not performed on the consonant portion and the background noise portion having relatively low power, and the vowel portion having high power is mainly waveform extended.

In the waveform expansion process 32, first, a process of extracting a pitch period (fundamental period) of a voice signal is performed, and two window functions having a length twice the extracted pitch period are input at two different input waveforms. Call. One of the window functions changes from 0 to 1, and the other changes from 1 to 0. By adding the results obtained by applying the window function to each other, a synthesized waveform having a two-pitch length is obtained. By inserting this synthesized waveform into the input waveform, it is possible to extend the waveform in the time axis direction without changing the pitch period.

Further, in the power judgment processing 32, the second threshold value To is compared with the power for each frame, and when the section having the power equal to or lower than To continues beyond the time threshold value, this section is used. Is regarded as a silent section, and the data in the output buffer corresponding to the portion exceeding the time threshold is deleted. As a result, the silent section in the output buffer is deleted, and the delay time caused by the speech speed conversion is shortened. In practice, in the power determination process 31, after a certain number of frames having power lower than the threshold To continue, the transfer to the output buffer is stopped until the next frame with power higher than To arrives. , This silence section deletion is realized.

[0013]

Generally, in hearing a hearing-impaired person, in addition to the problem that it is difficult to hear the quick speech, there is the problem that the weak consonant part cannot be heard well. However, in the above-mentioned conventional technique, only the method of slowing the voice by expanding the voice waveform is shown, and at the same time, the consonant portion of the voice is emphasized to make the slow voice more audible. Method was not considered.

An object of the present invention is to provide a concrete method for performing consonant emphasis processing which can be executed simultaneously with the conventional speech speed conversion processing.

[0015]

Means for achieving the above object of the present invention is to provide a consonant part from a time-series pattern of power values of input voice data in a voice speed conversion process of a conventional voice speed converter. A consonant detection means for detecting and a consonant emphasizing means for amplifying the value of the voice data of the consonant portion in the detected frame are provided.

[0016]

In the means for achieving the above object of the present invention, the consonant detection means observes a time series of power values of the voice data in the input frame as one pattern, and determines the predetermined power. The consonant part in the input frame is detected by the matching process with the pattern. The consonant emphasizing means partially amplifies the value of the audio data in the frame based on the result of the consonant detection to emphasize the consonant part.

[0017]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to embodiments.

FIG. 1 shows an embodiment of the speech speed conversion apparatus of the present invention. The present embodiment is different from the conventional voice speed conversion apparatus in that a consonant detection process 33 and a consonant emphasis process 34 are provided after the power determination process in the voice speed conversion process. As described in the section of the conventional technique, the power determination process is
The power of the entire frame determines the frame to which the waveform expansion process should be applied.

In general, when the power of the entire frame is large, this frame is often occupied by the vowel part of speech. On the contrary, when the power of the entire frame is small, it is considered that the frame is occupied by the consonant part and the silent part of the voice. In this embodiment, the threshold Th
The frame data having a larger power is subjected to the waveform expansion processing as in the conventional speech speed conversion device. And
When the power of the entire frame is smaller than Th, the consonant detection means sets the time series of the power values of the data in the frame to 1
It is regarded as one pattern and is compared with a predetermined power pattern. And as a result of the collation,
In the frame data determined to be a frame including a consonant, the value of a part of the data is amplified by the consonant emphasis process.

FIG. 2 shows a signal flow in the speech speed conversion processing unit of the present invention, and FIG. 3 shows a schematic diagram of consonant emphasis processing. In the consonant detection processing, the data of the input frame is divided into a plurality of smaller frames, and the power value for each small frame is calculated. 2 and 3 show the case where the frame is divided into four small frames. The power values of the four small frames are compared with the silence threshold To / 4, and "1" is set for the small frames having a power value larger than To / 4.
Small frames with power values less than To / 4 are labeled "0".

Next, the time series of these four labels is regarded as one pattern, and when the label time series is "0001" or "0011", consonants are included in the previous two small frames. Then, it is determined that the rear two small frames include the transition part to the vowel. Then, in the consonant emphasis processing, amplification processing is applied to the previous two small frame portions. The amplification processing does not uniformly amplify the two small frames, but rather gradually amplifies the small frames 2 as shown in FIG.
It is realized by multiplying the data of two corresponding small frames by an amplification coefficient string that gradually increases and decreases by one. This prevents the value from changing abruptly at the border of a small frame and causing noise.

On the other hand, if the time-series pattern of the label is a pattern other than the above, the consonant emphasis processing is not performed and the label is transferred to the output buffer as it is.

In the conventional speech speed conversion apparatus, a frame in which the power of the entire input frame is smaller than Th is output without any processing. However, according to the present invention, among the frames that have been output as they are until now, regarding the frames that are considered to include the consonant part of the voice,
The power of the consonant part is emphasized, and the consonant emphasis process can be realized simultaneously with the speech speed conversion process.

In the description using the above embodiment, the power value for determining the consonant portion is chronologically divided by 1/4 of the input frame.
Although the power of the unit is selected, the same effect can be obtained even if the same determination is performed by the time series of the power value in a finer unit.

Further, although the triangular coefficient sequence shown in FIG. 2 was used for the amplification processing for emphasizing the consonants, the same result can be obtained by using other amplification patterns, for example, a Gaussian function-shaped pattern. The effect is obtained.

Further, in the above embodiment, when labeling the power value of the small frame, To / 4 which is 1/4 of the silence threshold To used for determining the silence interval to be deleted.
However, other values may be used for labeling the small frames.

[0027]

According to the present invention, there is an effect that the consonant portion of the voice can be emphasized in parallel with the waveform expansion processing for the speech speed conversion.

[Brief description of the drawings]

FIG. 1 is a diagram showing a flow of signal processing according to an embodiment of the present invention.

FIG. 2 is a diagram showing a flow of signal processing of a speech speed conversion processing unit of the present invention.

FIG. 3 is a schematic diagram of consonant emphasis processing according to the present invention.

FIG. 4 is a diagram showing a flow of signal processing in a conventional speech speed conversion device.

FIG. 5 is an explanatory diagram of a voice speed conversion process in the voice speed conversion device.

[Explanation of symbols]

10 ... Input analog system, 11 ... Microphone, 12 ... Analog amplifier, 13 ... Low pass filter, 14 ... A / D converter, 20 ... Input ring buffer, 21, 22 ... Input frame buffer, 30 ... Speech speed conversion process, 31 ... power determination processing, 32 ... waveform expansion processing, 33 ... consonant determination processing, 3
4 ... Consonant enhancement processing, 40 ... Switch means, 41 ... Slow button, 50 ... Output ring buffer, 51, 52 ... Output frame buffer, 60 ... Output analog system, 61 ... Speaker or earphone, 62 ... Analog amplifier, 63 ...
Low pass filter, 64 ... D / A converter.

Claims (4)

[Claims] 1. A speech speed conversion apparatus for converting only speed without changing the pitch of a speech, comprising means for simultaneously performing processing for emphasizing a consonant portion of speech at the same time as speech speed conversion processing. And the speech speed converter. 2. The speed is converted without changing the voice pitch,
Further, in the speech speed conversion device having a unit for selecting a frame to be subjected to a process for expanding or contracting the time length of the waveform and a frame not to be subjected to the process by comparing the power of the input voice signal in units of frames and a threshold value, A consonant detection unit that determines whether or not a consonant portion is included in a frame that is not subjected to waveform expansion / contraction processing, and a consonant emphasizing unit that amplifies partial data of the frame when the consonant portion is included. A speech speed conversion device characterized by the above. 3. The consonant detection means further divides into a plurality of small frames a frame which is determined not to be subjected to the process of expanding or contracting the time length of the waveform by comparing the power of each frame with a threshold value, It is characterized in that whether or not a part of the frame is to be amplified and which part is to be amplified are determined by collating a time-series pattern of continuous power values for each small frame with a predetermined pattern. The speech speed conversion device according to claim 2. 4. The time series pattern of power values for each small frame is obtained by comparing the power value for each small frame with a second threshold value different from the threshold value.
The speech speed conversion device according to claim 3, wherein the speech speed conversion device is a binary time series pattern.