JPH10124098A - Speech processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To easily compress and restore a speech without deforming the information that a speech has. SOLUTION: A speech compressing device discriminates respective syllable data 'ha', 'na', 'ga', 'sa', 'i' and 'ta' by a syllable discriminating means 2 from, for example, speech data 'hanagasaita'. Then a vowel part detecting means 3 detects, for example, vowel part data 'A(a)' from syllable data 'ha(HA)', i.e., detect vowel part data from respective discriminated syllable data. Than a compressing means 7 makes the length of the corresponding data short by shortening the duration of each detected vowel part data to compress the speech on the whole. Further, a speech reproducing device restores the original speech data by inserting dummy waveform data based upon length information indicating the duration of a vowel part in the speech before compression added to the speech data to be restored into vowel part data in speech data to be restored by an expanding means.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio processing apparatus for processing audio, and more particularly, to an audio compression apparatus for compressing audio data by compressing vowel partial data constituting audio data, and a method for compressing vowel partial data. The present invention relates to an audio reproducing apparatus for restoring original audio data from reproduced audio data.

[0002]

2. Description of the Related Art It is common practice to transmit audio, record and store voice, and so on. For example, voice transmission is performed in a telephone device, and recording and storage of voice is performed in a tape recorder, for example. Have been done. In addition, as a method of performing audio transmission and recording / storage, in addition to the conventional analog method of processing audio as an analog signal, in recent years, a digital method of converting audio into a digital signal and then processing the digital signal has been generalized. With the development of digital signal processing technology, it has become common to process audio signals on digital computers.

[0003] In the case of voice transmission or recording / storage using either the analog system or the digital system, the transmission speed, the recording / storage efficiency and the like are reduced by compressing the audio to be transmitted, recorded and stored. Can be improved. For this reason, audio compression processing may be performed at the time of audio transmission or recording / storage, etc. As a method for compressing audio, a method using a frequency distribution (spectrum) or a distribution or correlation of a signal waveform is used. A method of using the method, a method of using a feature of a sound signal generation source and a sound source, and the like have been performed.

[0004]

However, in the above-described audio compression method, the compression processing method is complicated, and when the original audio is reproduced from the compressed audio data, the compression processing is not supported. There is a problem that it is necessary to perform complicated restoration processing. Further, in the conventional audio compression method, generally, the user cannot understand the contents of the audio information unless the compressed audio is expanded. For this reason, when using the compressed audio data, there is a problem that a device for decompression processing is always required.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned conventional problems, and compresses and decompresses speech by a novel method of directly compressing and decompressing phonemes constituting syllables in speech. An object of the present invention is to provide an audio processing device that can be easily performed. Another object of the present invention is to provide an audio processing device capable of performing audio compression without destroying information contained in the audio. Also,
An object of the present invention is to provide an audio processing device capable of compressing audio so that a reproduced audio can be grasped by a human even when compressed audio data is reproduced without being restored. With the goal.

[0006]

In order to achieve the above object, a voice compression apparatus for performing voice compression processing according to the present invention compresses voice as follows. That is, the syllable identification means converts, for example, each syllable data from the voice data “Hanagashita”, that is, “ha”, “na”,
Identify "ga", "sa", "i", "ta".

Next, vowel part detection means detects vowel part data from each of the identified syllable data. That is,
For example, the syllable data “wa (HA)” is composed of consonant part data “H (wa row)” and vowel part data “A (A)”, and from these data the vowel part detection means uses the vowel part data “A ( A) "is detected. Since the syllable data "I (I)" is composed of only the vowel part data "I (I)", the vowel part detecting means detects the vowel part data "I (I)". Then, the compression means shortens the duration of each of the detected vowel part data to shorten the length of the data, and compresses the voice as a whole. Note that, in order to faithfully restore the original voice by the reproduction process described later, length information indicating the length of the original vowel part data is added to the compressed voice data.

Therefore, the voice can be easily compressed by directly compressing the vowel parts constituting the syllables in the voice. In addition, it is possible to perform compression without destroying the information that the original audio has, that is, the information of “Hanagashita” in the above case, thereby reproducing the compressed audio data without restoring it. Even if you did, the original voice information, that is, in the above case,
The same effect can be obtained as in the case where the information is heard with a quick voice.

Further, in the audio reproducing apparatus for performing the audio reproducing process according to the present invention, the vowel part data constituting the syllable is compressed by shortening its length, and the vowel part data is compressed to the original length (ie, the original length). The original sound is reproduced as follows from the sound data to which the length information indicating the duration of the vowel in the sound is added. In other words, the syllable identification means converts, for example, each syllable data, that is, "ha", "na", "ga", "sa", "i""
Identify "ta". Next, vowel part detection means detects vowel part data from each identified syllable data. That is, for example, the syllable data “wa (HA)” is composed of consonant part data “H (wa row)” and vowel part data “A (a)”, and from these data the vowel part detection means uses the vowel part data “ A (A) "is detected. Since the syllable data "I (I)" is composed of only the vowel part data "I (I)", the vowel part detecting means detects the vowel part data "I (I)". And the extension means,
By inserting pseudo-waveform data having a length based on the length information into the detected vowel part data and restoring the duration of the vowel part data to the original length, the compressed voice data is expanded and expanded. Play the sound of.

Therefore, the vowel part data length of the compressed audio data can be restored to the original length, and the restored audio data can be reproduced with a sound quality close to the original audio. Here, the pseudo waveform data inserted into the vowel part data of the audio data to be restored may be the same or similar to the vowel part data of the original audio data. For example, in the audio data to be restored, Waveform data obtained by copying the vowel part data, or waveform data created in advance for each vowel by a real voice or a synthetic sound or the like can be used.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment according to the present invention will be described with reference to the drawings. First, an audio compression device that performs audio compression processing according to the present invention will be described. FIG. 1 shows an example of a voice compression device. The voice compression device includes a voice input unit 1 for inputting voice, a syllable partitioning unit 2 for separating voice data into syllable data, and a vowel from syllable data. First vowel-consonant identification unit 3 for identifying partial data and consonant partial data
And a second consonant identification unit 4 for identifying a consonant corresponding to the consonant partial data, a second vowel identification unit 5 for identifying a vowel corresponding to the vowel partial data, and storing sample data corresponding to each vowel and each consonant. A memory 6, a compression unit 7 for compressing vowel part data, a vowel / consonant synchronization circuit 8 for synchronizing vowel part data and consonant part data, and an output unit 9 for outputting voice data. In this example, with the above configuration, voice data is compressed by detecting vowel part data constituting each syllable data in the voice data and shortening the duration of each detected vowel part data.

The voice input unit 1 inputs voice by voice input means such as a microphone, for example, and outputs the input voice to the syllable delimiter 2 as voice data. The syllable separating unit 2 includes syllable identifying means for identifying each syllable data from the voice data, divides the voice data input from the voice input unit 1 into each syllable data, and separates each divided syllable data into a first vowel and consonant. Output to section 3. That is, for example, when voice data “Hanagashita” is input from the voice input unit 1, the voice separation unit 2 converts the input voice data into each syllable data, that is, “ha”, “na”, “ ",""Sa,"
Each syllable data divided into “i” and “ta” is output to the first vowel / vowel identification unit 3.

The first vowel-consonant discriminating unit 3 includes vowel part detecting means for detecting vowel part data from syllable data, and detects each syllable part data from each syllable data input from the syllable delimiter unit 2. The data is separated into consonant part data and vowel part data. The separated consonant part data is output to the second consonant identification unit 4, and the separated vowel part data is output to the second vowel identification unit 5. That is, for example, for the input syllable data “wa (HA)”, the first vowel-consonant identification unit 3 sets the syllable data “wa (H)”.
A) is divided into consonant partial data “H (ha row)” and vowel partial data “A (a)”, and the separated consonant partial data “H (ha row)” is And outputs the separated vowel part data “A (A)” to the second vowel identification unit 5.

Generally, a syllable constituting a voice is composed of a vowel part and a consonant part. For example, the syllable data "I (I)" is composed of only the vowel part data "I (I)". When the syllable data is composed only of the vowel part data, the first vowel / vowel identification unit 3 detects that the input syllable data is composed only of the vowel part data. The detected vowel part data, that is, the syllable data itself is used as the second vowel identification unit 5.
Output to Here, the method of identifying the vowel part data and the consonant part data performed by the first vowel / consonant identification unit 3 is not particularly limited. For example, as shown in FIG. Since it can be said that the amplitude is generally larger than the amplitude of the vowel part data, the vowel part data and the consonant part data can be distinguished by using the magnitude of the amplitude.

The second consonant discriminating section 4 discriminates a consonant corresponding to the consonant partial data input from the first vowel consonant discriminating section 3 and confirms that there is a consonant corresponding to the input consonant partial data. And outputs the input consonant partial data to the vowel-consonant synchronization circuit 8. That is, the second consonant identification unit 4
Refers to sample data corresponding to each consonant, such as "K (ka row)", "S (sa row)", or "T (ta row)" stored in the consonant memory 61 provided in the memory 6. For example, when the consonant partial data “H (wa row)” is input, the input consonant partial data and the consonant memory 6
1 is compared with the sample data corresponding to each of the consonants, the input consonant partial data is set to “H”.
("Line"), and outputs the consonant partial data "H (" line ") to the vowel-consonant synchronization circuit 8.

The second vowel identification section 5 identifies a vowel corresponding to the vowel partial data input from the first vowel / consonant identification section 3 and confirms that there is a vowel corresponding to the input vowel partial data. , And outputs the input vowel part data to the compression unit 7. That is, the second vowel identification unit 5 stores, for example, “A” stored in the vowel memory 62 provided in the memory 6.
(A) ”,“ I (I) ”, and“ U (U) ”, refer to the sample data corresponding to each vowel, and if, for example, vowel partial data“ A (A) ”is input, By comparing the input vowel part data with the sample data corresponding to each vowel stored in the vowel memory 62, it is determined that the input vowel part data is "A (A)", The vowel part data “A (A)” is output to the compression unit 7.

Each syllable data is stored in the first vowel-consonant identification unit 3.
When it is ensured that the vowel part data and the consonant part data are correctly separated by
The consonant discriminating unit 4, the second vowel discriminating unit 5, and the memory 6 do not necessarily need to be provided in the present configuration, but in this example, the second consonant discriminating unit 4, the second vowel discriminating unit 5, and the memory 6 are With the provision of this configuration, it is confirmed whether each syllable data is correctly separated by the first vowel / consonant identification unit 3 into vowel part data and consonant part data.

The compression unit 7 includes compression means for shortening the length of the vowel part data, and performs compression by reducing the length of the vowel part data input from the second vowel identification unit 5, that is, the duration. , And outputs the compressed vowel part data to the vowel-consonant synchronization circuit 8. Here, the method of shortening the duration of the vowel part data performed by the compression unit 7 is arbitrary. For example, since the vowel part data generally has a feature that periodic waveform data repeatedly appears,
For example, by cutting out a data portion composed of periodic waveform data from the vowel portion data shown in FIG. 2A as a compressed portion and deleting it, the duration of the vowel portion data is reduced as shown in FIG. You can also. In addition,
When such compression is performed, the original vowel part is easily inserted by inserting waveform data having the same or similar periodic waveform as the deleted part into the shortened vowel part data, as described later. Data can be faithfully restored.

Further, length information indicating the length of the original vowel part data, that is, the duration of the vowel part in the original voice, may be added to the compressed vowel part data. The audio reproducing apparatus for performing the audio reproducing process according to the present invention expands the duration of the compressed vowel part data based on the added length information and restores the duration of the original vowel part data. it can. The length information does not necessarily need to be added to each compressed vowel part data. For example, the length information of each vowel part data is collectively added to the head or end of the compressed audio data. You can also.

The vowel consonant synchronizing circuit 8 synchronizes the consonant partial data input from the second consonant discriminating section 4 with the compressed vowel partial data input from the compressing section 7 so that syllables corresponding to these data are obtained. The data is re-synthesized, the compressed speech data is re-generated from the re-synthesized compressed syllable data, and output to the output unit 9. That is, for example, the consonant partial data “H (is a row)” is input from the second consonant identification unit 4, and the compressed vowel partial data “A
(A) ”is input, the vowel-consonant synchronization circuit 8
The syllable data “wa (HA)” after compression is re-synthesized from these data, and for example, the syllable data “na”, “ga”, “sa”, “sai”, and “ta” after compression which are similarly re-synthesized. , And regenerates the compressed voice data “Hanagashita” from the output unit 9.
Output to The output unit 9 outputs the compressed audio data input from the vowel / consonant synchronization circuit 8 to, for example, a transmission path or a recording storage medium.

With the above configuration, the voice compression apparatus according to the present invention can easily perform voice compression by a novel compression method of shortening the duration of a vowel part constituting a syllable in the voice. Here, since the duration of a vowel part in a voice is generally longer than the duration of a consonant part, the duration of the vowel part in the voice is shortened to compress the voice into voice data having a relatively short duration. be able to. Also, the compressed audio data is compressed by the present invention because the duration of the vowel part data is reduced, and the audio is compressed without destroying the information that the original voice had. Even in the case of reproducing without restoring, the reproduced sound can be grasped by a human. In other words, even when the compressed audio data is reproduced without being restored, the same audio as when the original audio is uttered at a high speed is reproduced, so that the contents of the reproduced audio can be understood by a human. can do.

Next, a description will be given of an audio reproducing apparatus for performing an audio reproducing process according to the present invention. FIG. 3 shows an example of the audio reproducing apparatus. The audio reproducing apparatus includes an audio data input unit 11 for inputting audio data, and a consonant unit for determining a consonant corresponding to the consonant partial data in the audio data. A determining unit 12, a vowel determining unit 13 for determining a vowel corresponding to vowel partial data in the voice data, a memory 14 storing sample data corresponding to each vowel and each consonant, a length of the vowel partial data, That is, a vowel expansion unit 15 for restoring the duration and an output unit 16 for outputting audio data are provided. In this example, according to the above configuration, the length of the vowel part data constituting the syllable (that is, the duration of the vowel in the original voice) is reduced and the original length of the vowel part data is reduced. The original sound is reproduced from the sound data to which the length information indicating is added.

The audio data input unit 11 inputs compressed audio data via, for example, a transmission path or a recording storage medium, and outputs the input audio data to the consonant unit determination unit 12. The consonant part determination unit 12 includes syllable identification means for identifying each syllable data from the voice data, and vowel part detection means for detecting vowel part data from the syllable data, and converts the voice data input from the voice data input unit 11 Each syllable data is separated into vowel part data and consonant part data by dividing into each syllable data and detecting vowel part data from each divided syllable data. Then, after determining the consonant corresponding to each separated consonant partial data, each separated vowel partial data and each consonant partial data is determined by the vowel part determining unit 1.
Output to 3.

That is, when voice data such as “Hanagashita” is input, for example, the consonant part determination unit 12 converts the input voice data into each syllable data “ha”,
It is divided into "na", "ga", "sa", "i", and "ta". For example, for the syllable data "ha (HA)", the consonant partial data "H (wa row)" and the vowel partial data "A ( A) ". The consonant memory 1 provided in the memory 14
For example, by referring to the sample data corresponding to each consonant such as “K (ka row)”, “S (sa row)” or “T (ta row)” stored in 41, for example, the separated consonant partial data “ By comparing the consonant partial data with the sample data corresponding to each consonant stored in the consonant memory 141, the separated consonant partial data becomes “H (wa row)”. It is determined that there is. And
The separated vowel part data and consonant part data are output to the vowel part determination unit 13.

As described above, for syllable data composed only of vowel part data, the consonant part determining unit 12 does not judge the consonant part data, and uses the syllable data itself as vowel part data. Judgment unit 13
Output to Here, there is no particular limitation on a method of identifying both data in the separation processing of vowel part data and consonant part data performed by the consonant part determination unit 12, and for example,
Similarly to the above, the vowel part data and the consonant part data can be identified and separated using the magnitude of the amplitude.

After determining the vowel corresponding to the vowel part data input from the consonant part determination section 12, the vowel part determination section 13 outputs the input vowel part data and consonant part data to the vowel expansion section 15. That is, the vowel part determining unit 1
For example, when the vowel partial data “A (A)” and the consonant partial data “H (ha row)” are input, the memory 1
4 with reference to the sample data corresponding to each vowel such as “A (A)”, “I (I)”, or “U (U)” stored in the vowel memory 142 provided in the vowel. By comparing the partial data with the sample data corresponding to each vowel stored in the vowel memory 142, it is determined that the input vowel partial data is “A”. Then, the input vowel part data “A
(A) ”and consonant partial data“ H (ha row) ”are output to the vowel expansion unit 15.

By the above-described processing for determining consonant part data and vowel part data in the consonant part judging section 12 and the vowel part judging section 13, it is possible to confirm whether the consonant part judging section 12 has correctly separated the two data. it can. Also,
The determination process of the vowel part data and the consonant part data does not necessarily need to be performed. For example, when a vowel corresponding to the vowel part data is determined, the vowel part data At the time of the restoration processing of the length, pseudo waveform data corresponding to each vowel prepared in advance can be used in correspondence with the determined vowel.
In the present example, the consonant part determining unit 12 separates the voice data into syllable data, and further separates the separated syllable data into vowel part data and consonant part data to determine consonant part data. Later, the vowel part data was determined by the vowel part determination unit 13. However, the order of these determination processes is arbitrary. In short, the vowel part data and the consonant part data are separated from each syllable data constituting the voice data. I just want to be able.

The vowel expansion section 15 includes expansion means for restoring the length of the vowel section data. The length of the vowel section data input from the vowel section determination section 13, ie, the length information to which the duration is added. Restore based on Then, the original syllable data is re-synthesized from the restored vowel part data and the input consonant part data, the original speech data is re-generated from the re-synthesized syllable data, and output to the output unit 16.
That is, the vowel expansion unit 15 copies, for example, a part (vowel base) of the vowel partial data to be restored shown in FIG. 4A based on the length information added to the vowel partial data to be restored. By inserting the pseudo-waveform data into the vowel part data to be restored (vowel reproducing unit), the same length as the original vowel part data as shown in FIG.
That is, the vowel part data having the duration is restored.
Then, for example, the restored vowel partial data “A (A)”
And the consonant partial data “H (wa row)” input and the syllable data “ha (H
A)), and regenerates the original voice data "Hanagisaita" from the syllable data "na", "ga", "sa", "i", "ta" which are similarly resynthesized, for example. And outputs it to the output unit 16.

Although the data portion corresponding to the length information is omitted in FIG. 4 (a), the length information may be, for example, in each compressed vowel part data or in the decompressed data. It is added to the beginning or end of the target audio data. The output unit 16 reproduces and outputs the restored audio data input from the vowel expansion unit 15 by audio output means such as a speaker, or records and accumulates the audio data by audio recording and accumulating means such as a tape recorder. With the configuration described above, in the audio reproduction device according to the present invention, it is possible to restore the original vowel partial data length, that is, audio data composed of the duration, from the audio data in which the length of the vowel partial data is reduced, It is possible to reproduce the sound data restored with sound quality close to the original sound.

In this embodiment, the pseudo waveform data obtained by copying the vowel part data in the audio data to be restored is used as the pseudo waveform data to be inserted into the vowel part data of the audio data to be restored. As the pseudo waveform data, what kind of waveform data is the same or similar to the vowel part data of the original voice data, for example, waveform data created in advance for each vowel is prepared. Waveform data may be used.

Further, the audio compression device and the audio reproduction device according to the present invention may be used separately, and when these devices are used in combination, for example, the communication between the audio compression device and the audio reproduction device may be performed. It is also possible to compress the audio data for transmission. Further, the audio compression processing and the audio reproduction processing according to the present invention can be performed in combination with the conventional audio compression processing and the audio reproduction processing. For example, by performing the audio compression processing according to the present invention as a preprocessing of the conventional audio compression processing, Thus, the audio can be compressed at a higher compression ratio than when only the conventional compression processing is performed.

[0032]

As described above, according to the speech processing apparatus of the present invention, the compression and restoration of speech can be easily performed by a novel method of directly compressing and restoring phonemes constituting syllables in speech. It can be carried out. In addition, audio compression can be performed without destroying information contained in the audio,
Thus, even if the compressed audio data is reproduced without being restored, the audio can be compressed so that the reproduced audio can be grasped by a human.

[Brief description of the drawings]

FIG. 1 is a configuration example of an audio compression device according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining audio compression processing according to an embodiment of the present invention.

FIG. 3 is a configuration example of an audio reproduction device according to an embodiment of the present invention.

FIG. 4 is a diagram for explaining audio reproduction processing according to one embodiment of the present invention.

[Explanation of symbols]

1. Voice input section 2. Syllable section section 3. First
Vowel-consonant discriminator, 4 second consonant discriminator, 5 second vowel discriminator, 6 memory, 7 compression unit, 8 vowel-consonant synchronization circuit, 9 output unit, 61・ Consonant memory,
62 vowel memory, 11 voice data input section
12 ··· Consonant part judgment unit, 13 ··· Vowel part judgment unit, 14
..Memory, 15..vowel expansion unit, 16..output unit, 141..consonant memory, 142..vowel memory,

Claims (2)

[Claims] 1. A voice compression apparatus for compressing voice, comprising: syllable identification means for identifying each syllable data from voice data; vowel part detection means for detecting vowel part data from the identified syllable data; Compression means for compressing the audio data by shortening the length of the partial data. 2. An original speech is reproduced from speech data compressed by shortening the length of vowel part data constituting a syllable and having length information indicating the original length added to the vowel part data. In the sound reproducing apparatus, syllable identification means for identifying each syllable data from the audio data, vowel part detection means for detecting vowel part data from the identified syllable data, and the detected vowel part data based on the length information And a decompression means for restoring the original audio data by inserting pseudo-waveform data of a length.