JPH07334189A - Sound information analysis device - Google Patents

Abstract

PURPOSE:To remove a dynamic noise mixed and added to an input sound signal and to improve the S/N ratio of the sound signal by extracting a noise characteristic from the frame data incorporating no sound excepting a noise, using the extracted updated noise characteristic and removing the noise. CONSTITUTION:This device is provided with a sound/silence decision part 500 and an analysis part 600. The analysis part 600 is provided with an FFT part 660 for obtaining a pitch period from a logarithmic value of a power spectrum, a pitch extraction part 670 extracting the characteristic and the repeat period (pitch information) of the sound, a noise process part 720 extracting the noise characteristic, a frame setting part 730 for silence setting the data for outputting the silence and a table 710. Then, the noise process part 720 stores noise information at a silence time in the table 710. In a noise removal part 680, noise removal and a process improving the S/N of the sound are performed by a spectrum line at a sound time by using a frequency characteristic information value stored in the table 710.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice information analysis device for analyzing voice information, and more particularly to a voice information analysis device for improving the S / N ratio with respect to noise that dynamically changes in a voice signal.

[0002]

2. Description of the Related Art As a conventional technique for satisfactorily transmitting an input voice signal, there is Japanese Patent Application No. 5-138626.

The technique disclosed in Japanese Patent Application No. 5-138626 is as follows.
By using a noise table for noise removal, the information extracted when there is no sound is treated as noise information, and the information in the noise table is referenced when there is sound, and noise is removed.
Good voice signal was transmitted.

[0004]

DISCLOSURE OF THE INVENTION The present invention, in the above-mentioned device, is not affected by any use application or use environment,
In addition, the voice information analysis and noise removal effectively removes the dynamic noise that is mixed and added to the input voice signal without increasing the hardware limitation, and improves the S / N ratio of the voice signal to obtain good voice. An object is to provide a high-quality voice information analysis device for the purpose of transmission.

[0005]

[Means for Solving the Problems] To achieve the above object,
The present invention is a voice information analysis method for outputting an analysis result obtained by analyzing voices represented by frame data obtained by accumulating voice sampling data for a certain period of time, and voices other than noise are included in the voices indicated by each frame data. And a step of determining whether or not there is a voice other than noise, and when it is determined that the voice included in the voice indicated by the frame data is extracted from the frame data and stored, As a result of analysis of the voice indicated by the data, a step of executing a silent system process of outputting information indicating a result of analysis of a silent voice prepared in advance, a means for extracting and storing characteristic information of the voice from pitch information, and noise. When it is determined that the sound other than the noise is included, the noise feature stored when it is determined that the sound other than the noise is not included And removing the voice signal indicated by the voice data from the voice data, emphasizing the voice signal based on the feature information of the voice obtained from the pitch information, and outputting the analyzed result. A voice information analysis method is provided.

[0006]

According to the voice information analysis method of the present invention, it is determined whether the voice indicated by each frame data contains voice other than noise, and it is determined that voice other than noise is not included. In this case, the characteristics of noise included in the voice indicated by the frame data are extracted from the frame data and stored, and the information indicating the analysis result of the silent voice prepared in advance is obtained as the analysis result of the voice indicated by the frame data. Silence processing that outputs is executed.

On the other hand, when it is determined that a voice other than noise is included, the noise feature stored in the previous silence processing is removed from the voice indicated by the frame data, and the pitch period, which is a voice feature, is removed. And a process of emphasizing a frequency that is an integral multiple thereof are executed, and a sound system process for outputting an analysis result of frame data with an improved S / N ratio is executed.

Therefore, since the latest noise feature is always extracted, the noise is removed by using the extracted latest noise feature during the period in which voice other than noise is included. Since the latest pitch information is always obtained and the voice feature information is stored, it is possible to immediately respond to a change in the speaker in use and emphasize the voice feature of the corresponding speaker.

Further, since voiced system processing and silent system processing do not occur at the same time, the execution load of these processes is small and can be realized on a single processor without limiting the voice information analysis function. .

[0010]

EXAMPLES Some examples of the present invention will be described below.

FIG. 4 shows the configuration of a communication system to which the voice information analyzing apparatus according to the present invention is applied.

In the figure, 1000 is a transmitter and 2000 is a receiver.

The transmitter 1000 compresses and encodes a voice signal by a method utilizing voice analysis, and obtains level information 7
01 and pitch information 671 are transmitted to the receiving device 2000.

The transmitting apparatus 1000 has a voice input unit 200 which is a voice input unit, and an A / D converter which converts the input voice input thereto into analog / digital (A / D) and converts it into voice sampled data. It has a container 300, a buffer memory 400 for sequentially storing the voice sampling data, and a voice / non-voice determining unit 500 for determining voice / non-voice.

The buffer memory 400 has a fixed time (10
(About 30 milliseconds), the sound / silence determination unit 50 stores the sound sampling data (frame data) 401 for a certain time.
Send to 0.

Further, the transmitting apparatus 1000, from the frame data 401, the normalized waveform sequence 702, the level information 7
01, and an analysis unit 600 that creates pitch information 671,
It has a vector quantizer 800 for converting the normalized waveform sequence 702 into a vector code 801, and a transmitter 900 for transmitting these pieces of information.

As a first embodiment, FIG. 1 shows blocks of a voiced / non-voiced determination unit 500 and an analysis unit 600 according to claim 1.

The voice / silent discriminator 511 output from the voice / silent discriminating unit 500 divides the process into a voice processing system and a silence processing system.

The analysis unit 600 includes an FFT data setting unit 610 for setting data for FFT and frame data 40.
F from 1 to means for obtaining frequency characteristics by the FFT
The FT unit 620 and the power spectrum conversion unit 63 that outputs the square value of the absolute value of the obtained complex number and the power spectrum.
0, an axis conversion unit 640 that converts the vertical axis from a power spectrum axis to an amplitude axis, a noise removal unit 680 that removes noise using the table 710, an inverse FFT unit 690 that is an inverse FFT unit, and an inverse FFT unit 690. A normalization unit 700 for setting the maximum value of the FFT result to “1”, an FFT unit 660 for obtaining a pitch period from the logarithmic value (cepstral) of the power spectrum, and a feature (pitch) of speech and repetition. Pitch extractor 670 for extracting the period (pitch information)
A noise processing unit 720 for extracting the characteristics of noise, a silence frame setting unit 730 for setting data for outputting silence, and a table 710.

The FFT (Fast Fourier)
Transform: Fast Fourier Transform) section 620, 6
60 is a technique for reproducing an original waveform from a sampled value of a signal by frequency and amplitude DFT (Discrete F
This is a faster implementation of the "ourier Transform" (discrete Fourier transform).

The FFT signal processing technology is described in "Introduction to Signal Processing" Yoshifumi Amemiya / Yukio Sato Chopsticks Ohmsha P106-
6.3 "Fast Fourier Transform" describes in detail.

As a method for obtaining the pitch period in the pitch extraction section 670, a parallel processing method, a data reduction method,
There are a cepstrum method, a period histogram method, and the like.
("Digital Signal Processing" by Sadahiro Furui) In the pitch extraction section 670, since the characteristics of the input speech signal appear above the integral multiple of the pitch cycle (fundamental frequency), it is generated based on the obtained pitch cycle. Information on the table 710
Remember.

Table 7 is an example of a method for removing noise and improving the S / N ratio using the above table 710.
A method of using "weighting" for the type of information stored in 10 can be considered.

The "weighting" is actually 0.0-1.0.
It is expressed by the numerical value of.

The table 710 is initialized to "1.0".

In the table 710, 0.0 to 1.0 for the power spectrum sequence (hereinafter power spectrum).
There is a method in which a value obtained by “weighting” is stored and a power spectrum sequence in the presence of a voice (hereinafter referred to as a voice power spectrum) is multiplied. The table 710 stores the result of weighting the noise power spectrum from 0.0 to 1.0 in descending order of absolute value.
That is, by multiplying the frequency in which noise is noticeable by 1.0 or less, the corresponding power spectrum value of the voiced power spectrum is shifted in the direction of decreasing from the original value, and the frequency at which the noise spectrum does not appear. Is weighted with 1.0, the power spectrum value of the frequency is left as it is. As a result, it is possible to obtain a power spectrum sequence from which noise has been removed.

The noise processing section 720 stores the noise information when there is no sound by the above method.

Furthermore, in contrast to the table 710 created in advance by the noise processing section 720, the pitch extraction section 670 stores information stored in the table 710 (already stores 0.0 to 1.0). ) Is added in the increasing direction (temporary addition value: 0.0 to 1.0). Therefore, since the value of the table 710 corresponding to the pitch period frequency, that is, the frequency that is an integral multiple of the fundamental frequency takes a value of 1.0 to 2.0, the corresponding frequency is emphasized more than the actual frequency and is reproduced. Sound signal is emphasized. At this time, the added value to the table 710 is required to be the same value for any frequency (an integral multiple of the fundamental frequency). If they are not the same value, the original audio information will be lost,
This is because the voice itself is processed.

In this way, the table area can be saved by using the noise information and the voice signal characteristic information in the same table.

In the noise removing section 680, the table 7
Using the frequency characteristic information value stored in 10, processing is performed to remove noise and increase the S / N of the voice from the spectrum sequence in the presence of voice.

Next, the second embodiment will be described.

FIG. 2 shows blocks of the voiced / silent part 500 and the analysis part 600 according to claim 1.

The second embodiment is similar to the first embodiment in that the voice /
Only the configurations of the silence determination unit 500 and the analysis unit 600 are different.

The sound / silence judging section 500 includes an FFT data setting section 610 for setting data for FFT processing of frame data, an FFT section 620 for FFTing frame data, and a square of the obtained complex number. A power spectrum conversion unit 630 for obtaining a sum and an audio power determination unit 51 for obtaining a sum of frame data from the frame data, comparing the sum with a threshold value, and outputting a sound / silence determiner 511.
0, a voice / silence determination unit 650 for determining the voice / silence determiner, an FFT unit 660 for FFT processing the power spectrum, logarithm of the power spectrum, and pitch information extracted from the cepstrum, The pitch period is not fixed during silence ("Digital signal processing")
Furui Sadahi chopsticks P57-P59 4.9 pitch extraction)
It has a pitch extraction unit 670 that determines whether there is sound or no sound and outputs a sound / soundless discriminator.

In determining the presence / absence of a sound, the pitch information on the time axis is converted into a pitch period on the frequency axis, and if there is a maximum value for each pitch period in the power spectrum, the presence or absence of the sound is present. For example, it realizes the method of judging as silence.

If the cepstrum method is used as a method for obtaining the pitch period (fundamental frequency), the envelope of the spectrum is obtained by separating the fine structure by the Fourier transform of the logarithm of the power spectrum. The obtained pitch period is stored in the table 710.

Further, the analysis unit 600 includes the axis conversion unit 64.
0, the noise removing unit 680, the inverse FFT unit 690,
Having the normalization unit 700, the voice / silence unit 740, the noise processing unit 720, and the silence data setting unit 730, when it is determined to be a silence processing system, all of the obtained information is noise data in the table 710. If it is determined to be a sound processing system, the noise removal unit 680 uses the table 710 to process.

The voice / silent discriminator 511 output from the voice / silent discriminating section 500 divides the process into a voice processing system and a silence processing system.

The analysis section 600 includes an axis conversion section 640 for converting the vertical axis from a power spectrum axis to an amplitude axis, a noise removal section 680 for removing noise, an inverse FFT section 690 which is means for inverse FFT, and an inverse FFT. The maximum value of the results
It has a normalization unit 700 for setting to 1 ″, a noise processing unit 720 for extracting noise characteristics, a silence data setting unit 730 for setting data for outputting silence, and a table 710.

As described above, the first embodiment and the second embodiment have the same configuration but the same number of processing steps (blocks). The second embodiment has the same processing time as the first embodiment,
The sound / silence processing system is selected more strictly. Therefore, it is possible to remove the noise and improve the S / N ratio, which are the objectives.

Next, a third embodiment will be described.

The third embodiment differs from the first and second embodiments only in the structure of the table used.

A block diagram of the analysis unit 600 according to claim 1 is shown in FIG.

In the third embodiment, the noise information and the voice characteristic information are managed in separate tables.

In the noise processing section 720, the spectrum of the frequency corresponding to the noise frequency is stored in the table 71.
0, while the pitch extraction unit 670 stores a table 750 of the corresponding frequencies. After that, in the noise removing unit 660, after the frequencies stored in the table 710 are subtracted, the purpose is realized by using the table 750.

As described above, a method of separately storing various information by using two or more tables can be considered.

As described above, according to this embodiment, the constantly changing ambient noise is always extracted, the noise feature is removed, the feature of the input voice signal is always extracted from the pitch information, and the feature is emphasized. As a result, it is possible to deal with any situation, to improve the S / N ratio accurately, and to perform high-quality audio signal processing.

By the way, the processing of each unit according to the first, second and third embodiments can be realized as a program which operates on a processor. In this case, when there is no sound, only the processing of the silent processing system needs to be performed, and when there is sound, only the processing of the voice processing system needs to be performed, and the pitch information calculated for the silent / sound determination or normalization is used. By doing
No new processing or device is required, the load on the processor is small, and it can be realized as a program that operates on a single processor.

In the above embodiments, the application of the communication system has been described as an example. However, in the voice information analysis apparatus according to the first, second and third embodiments, the analysis result of the analysis unit 600 is also included. A device that performs processing such as voice recognition using
It can be applied to various devices.

[0050]

The present invention not only performs accurate noise removal under any circumstances by constantly extracting ambient noise that constantly changes and removing its noise characteristics, but also always provides the latest pitch information (fundamental frequency). By using and to emphasize the voice frequency, it is possible to cope with the change of the speaker and to improve the S / N ratio.

Further, by increasing the conditions for selecting the voiced / silent system processing, by reliably creating the voice information table from the voice information, the effect can be enhanced in the voiced system processing.

Furthermore, at least one voice information table is required, and depending on the number of voice information tables, the effect suitable for the intended use can be realized without changing the hardware.

[Brief description of drawings]

FIG. 1 is a block diagram showing a sound / silence determination unit and an analysis unit according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a sound / silence determination unit and an analysis unit according to a second embodiment of the present invention.

FIG. 3 is a block diagram showing a sound / silence determination unit and an analysis unit according to a third embodiment of the present invention.

FIG. 4 is a block diagram showing a voice information device of the present invention.

[Explanation of symbols]

1000 ... Voice information analysis device (sending side) according to the present invention, 200 ... Voice input device, 300 ... A / D converter, 400 ... Buffer memory, 401 ... Frame data, 500 ... Voice / silence determination unit, 510 ... Voice power determination unit, 511 ... Voice / silence determiner, 600 ... Analysis unit, 610 ... FFT data setting unit, 620 ... FFT unit, 630 ... Power spectrum conversion unit, 631 ... Power spectrum sequence, 640 ... Axis conversion unit, 670 ... Pitch extraction section, 671 ... Pitch information, 680 ... Noise removal section, 690 ... Inverse FFT section, 700 ... Normalization section, 701 ... Level information, 702 ... Normalized waveform sequence, 710 ... Noise table, 720 ... Noise processing Section, 730 ... Silence frame setting section, 740 ... Spoken / Silence section, 750 ... Voice information table, 650 ... Spoken / Silence , 660 ... FFT section, 800 ... Vector quantization section, 801 ... Vector code, 900 ... Transmission section, 2000 ... Speech information analysis apparatus (reception side) according to claim 1, 2100 ... Reception section, 2200 ... Vector dequantization Unit, 2300 ... Synthesis unit, 2400 ... Buffer memory, 2500 ... D / A conversion unit, 2600 ... Audio output device.

Continuation of front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display area H04B 15/00

Claims (4)

[Claims] 1. A voice information analysis device for outputting a result of analysis of voice information represented by frame data, which is obtained by accumulating voice sampled data for a certain period of time, wherein the voice represented by each frame data includes voice other than noise. If it is determined that the sound other than noise is not included, the characteristics of the noise included in the sound indicated by the frame data are extracted from the frame data and stored in a table. Then
As a voice analysis result indicated by the frame data, a step of executing a silence system processing for outputting information representing the analysis result of a silence voice prepared in advance, and a table of voice pitch information when voices other than noise are included. When it is determined that the means for storing in and the sound other than noise is included,
The process of emphasizing the characteristics of the audio signal obtained from the pitch information, and when it is determined that the audio other than noise is not included, the characteristics of the noise stored in the table are removed from the audio represented by the frame data to remove the noise. And a step of analyzing a voice represented by the frame data from which the characteristic is removed and outputting a result of the analysis, the voice information analyzing apparatus. 2. A voice compression encoding apparatus comprising: the voice information analyzing apparatus according to claim 1; and means for quantizing the waveform information output by the voice information analyzing apparatus and outputting quantized data. . 3. The voice information analyzing apparatus according to claim 1 or the voice compression encoding apparatus according to claim 2, further comprising means for transmitting via a wired or wireless transmission path. Terminal device. 4. A communication system comprising one or a plurality of the communication terminal devices according to claim 3.