Voice information analyzing device

Description: TECHNICAL FIELD [Detailed description of the invention]

【0001】

[Industrial application field]

The present invention relates to a voice information analyzer that analyzes voice information, and more particularly to a technique for removing a dynamically changing noise component from a voice signal.

【0002】

[Conventional technology]

As a conventional technique for removing noise from an input voice signal, a technique described in Japanese Patent Application Laid-Open No. 2-278298 and a technique described in Japanese Patent Application Laid-Open No. 1-75593 are known.

【0003】

The technique described in JP-A-2-278298 is a technique in which a filter is subdivided and used to fix a specific frequency, and the sound quality depends on the frequency to be removed. Further, the technique described in Japanese Patent Application Laid-Open No. 1-75593 is a technique of learning / extracting the characteristics of noise in advance using a neural network for noise removal and removing the noise by using the neural network, and the sound quality thereof. Depends on the noise selection that mixes the noise characteristics into the data (voice + noise) used for learning / extraction.

【0004】

[Problems to be Solved by the Invention]

According to the technique for removing a specific frequency using the noise removing filter described in JP-A-2-278298, since the frequency to be removed is fixed, the speaker and the background are moved. It is not suitable for removing noise whose characteristics change dynamically with changes.

【0005】

Further, even with the technique of extracting / learning the noise characteristics described in JP-A-1-75593 in advance, it takes a long time to learn the noise due to the nature of the neural network. , Not suitable for removing noise whose characteristics change dynamically with changes.

【0006】

In addition, when performing voice information analysis, if these technologies are applied to remove noise from the voice signal to be analyzed for voice information, in any case, voice analysis processing is performed as preprocessing for voice information analysis processing. The noise removal process must be performed separately from. Therefore, when performing real-time voice information analysis processing, there is a possibility of overload, and it is realized by limiting the voice information analysis function or by executing each processing with a different processor. There may be a need.

【0007】

Therefore, an object of the present invention is to provide an audio information analysis device capable of efficiently removing noise whose characteristics change dynamically from an audio signal to be analyzed for audio information.

【0008】

[Means for solving problems]

In order to achieve the above object, the present invention is a voice information analysis method for outputting an analysis result obtained by analyzing a voice indicated by a frame data in which voice sampled data is accumulated for a certain period of time, and the voice represented by each frame data is used. The step of determining whether or not the sound other than noise is included, and the characteristics of the noise contained in the sound represented by the frame data from the frame data when it is determined that the sound other than noise is not included. Is extracted and stored, and as the analysis result of the voice indicated by the frame data, a silent process for outputting information representing the analysis result of the silent voice prepared in advance is executed, and the voice other than noise is included. If it is determined that, the noise feature memorized in the previous silent processing is removed from the voice represented by the frame data, and the voice indicated by the frame data from which the noise feature is removed is analyzed, and the analysis result is obtained. Provided is a voice information analysis method characterized by having a step of executing an output sound system process.

【0009】

[Action]

According to the voice information analysis method according to the present invention, it is determined whether or not the voice represented by each frame data contains voice other than noise, and when it is determined that the voice other than noise is not included. , The characteristics of the noise contained in the voice represented by the frame data are extracted from the frame data and stored, and as the analysis result of the voice indicated by the frame data, the information representing the analysis result of the silent voice prepared in advance is stored. Executes the silent system processing to be output. On the other hand, when it is determined that the sound other than the noise is included, the noise feature memorized in the previous silence processing is removed from the voice represented by the frame data, and the noise feature is removed. Analyzes the voice indicated by and executes sound processing that outputs the analysis result.

【0010】

Therefore, the latest noise feature extraction is always performed during the period when no noise other than noise is included, that is, during the period when it is sufficient to output the analysis result of silence obtained in advance by regarding it as silence. During the period when is included, noise can be removed by using the latest extracted noise characteristics. Further, since the sound system processing and the sound system processing do not occur at the same time, the execution load of this processing is small, and it can be realized on a single processor without limiting the voice information analysis function.

【0011】

[Example]

Hereinafter, an embodiment of the present invention will be described.

【0012】

First, the first embodiment will be described.

【0013】

FIG. 5 shows a configuration of a communication system to which the voice information analyzer according to the present invention is applied.

【0014】

In the figure, 2000 is a transmitting device and 1000 is a receiving device.

【0015】

The transmitting device 2000 transmits the level information, the quantization data, and the pitch information obtained by compressing and encoding the voice signal by a method using voice analysis to the receiving device 1000. The receiving device 1000 decodes the voice from the received information and outputs it.

【0016】

Here, the transmission device 200 includes a transmission unit 900, a vector quantization unit 800, and a voice information analyzer 100, and the reception device 1000 includes a reception unit 1100, a vector inverse quantization unit 1100, and a synthesis unit. It is equipped with 1200, a D / A converter 1300, a buffer memory 1500, and an audio output device 1400.

【0017】

In the transmission device 2000, the voice information analyzer 100 analyzes the input input voice and sends the obtained level information 681 and pitch information 651 to the transmission unit 900 and the normalized waveform sequence 682 to the vector quantization unit 800. send. The vector quantization unit 800 converts the received normalized waveform series 682 into a vector code, and sends the converted quantization data 801 to the transmission unit 900. The transmission unit 900 receives the level information 681, the pitch information 651, and the quantization data 801 and transmits them to the receiving device 1000 via wired / wireless.

【0018】

On the other hand, in the receiving device 1000, the receiving unit 1000 receives the information transmitted from the transmitting device 2000 and outputs the level information 681', the quantization data 801', and the pitch information 651'. The vector dequantization unit 1100 dequantizes the output quantization data 801'. The synthesis unit 1200 synthesizes the waveform by superimposing the waveform 682'output by the vector inverse quantization unit 1100 for each repetition period (pitch information) based on the pitch information 651', and stores the buffer memory 1500. .. The D / A conversion unit 1300 converts the output of the buffer memory 400 into digital / analog (D / A). The audio output device 1400 outputs the audio obtained by the D / A converter 1300.

【0019】

Hereinafter, the details of the voice information analyzer 100 will be described.

【0020】

As shown in FIG. 5, the voice information analyzer 100 converts the input input voice into analog / digital (A / D) conversion with the voice input unit 200, which is a voice input means, into voice sampling data. It has an A / D converter 300 for conversion and a buffer memory 400 for sequentially storing the voice sampling data. The voice sampling data for each fixed time (10 to 30 milliseconds) stored in the buffer memory 400 is sent to the sound / silence determination unit 500 as the fixed time voice sampling data (frame data) 401. Further, the voice information analyzer 100 has a sound / silence determination unit 500 that determines sound / silence, and the voice information analyzer 100 has a normalized waveform series 682, level information 681, and pitch information from the frame data 401. It has an analysis unit 600 that creates 651.

【0021】

First, the sound / silence determination unit 500 is composed of the voice power determination unit 510 as shown in FIG. The voice power determination unit 510 obtains the sum (power) of each element of the frame data 401, compares it with the threshold value (fixed value), and determines that there is no sound if it is smaller than the threshold value and that it is sound if it is larger than the threshold value. Outputs a sound / silence judge 511 that represents sound / silence.

【0022】

If the above threshold value does not provide a sufficient effect, the threshold value may be made variable according to Equation 1. In Equation 1, ω is a weight value that satisfies 0 <ω <1.

【0023】

Next frame data threshold = ω Threshold of previous frame data + (1-ω) Current silent power value .. (Equation 1) That is, when it is determined that there is no sound, the power value determined to be silent and the threshold value used for the determination of the previous frame data are appropriately weighted, and the added value of this is used for the determination of the next frame data. Try to find the threshold to use.

【0024】

Further, the sound / silence determination of the voice power determination unit 510 may be performed as follows. That is, the power of the frame data to be determined is compared with the power of the previous frame data, and if the difference is larger than the predetermined value, it is determined that the sound / silence state has changed from the previous time, and the difference is. If it is smaller than the predetermined value, it is determined that the sounded / silent state has not changed. Then, based on the determination result of the stored pre-frame data, the sound / silence determination of the determination target frame data is performed. Alternatively, when the difference is larger than a predetermined value, the power of the frame data to be determined is compared with the threshold value to determine whether there is sound or no sound.

【0025】

Next, as shown in FIG. 2, the analysis unit 600 sets data in the FFT unit 620 and the FFT unit 620 that obtain the frequency characteristics of the frame data from the frame data 401 by FFT (Fast Fourier Transform). It has an FFT data setting unit 610 that performs the above. Further, the square value of the absolute value of the complex number obtained by the FFT of the FFT unit 620, that is, the power spectrum transform unit 630 that outputs the power spectrum and the axis transform that converts the vertical axis of the power spectrum from the power spectrum axis to the amplitude axis. It has a part 640. FFT (Fast Fourier Transform) is a faster implementation of DFT (Discrete Fourier Transform), a technique that reproduces the original waveform from the sampled value of the signal by frequency and amplitude. Such signal processing technology by FFT is explained in detail in "Introduction to Signal Processing" Yoshifumi Amemiya / Yukio Sato Chopsticks Ohm Co., Ltd. P106 ~ 6.3 "Fast Fourier Transform".

【0026】

Further, the analysis unit 600 includes a sound processing system composed of an FFT unit 730, a pitch extraction unit 650, a noise removal unit 660, an inverse FFT unit 670, and a normalization unit 680, a noise processing unit 690, and a silent data setting unit. It has a silence processing system composed of 700, a noise table 710, and a sound / silence section 720.

【0027】

The sound / silence unit 720 determines the sound / silence judge 511 output from the sound / silence determination unit 500, and if there is sound, the sound processing system performs processing, and if there is no sound, the sound processing system performs processing. Let the silence processing system perform the processing.

【0028】

First, the operation of the silence processing system when the sound / silence unit 720 determines that there is no sound will be described.

【0029】

In this case, the noise processing unit 690 stores the noise characteristics in the noise table 710 from the output of the axis conversion unit 640. That is, for example, the frequency characteristic information and the power spectrum sequence axis-converted by the axis conversion unit are stored in the noise table 710. The silence frame setting unit 700 outputs the data 682b representing the silence as the normalized waveform sequence to the vector quantization unit 800 as the normalized waveform sequence 682.

【0030】

Next, the operation of the sound processing system when the sound / silence section 720 determines that the sound is sound will be described.

【0031】

In this case, the FFT unit 730 is used to obtain the pitch period from the logarithmic value (cepstrum) of the power spectrum. The pitch extraction unit 650 extracts voice characteristics (height) and repetition period (pitch information) from the output of the FFT unit 730. The noise removing unit 660 removes noise from the power spectrum sequence at the time of sound output by the axis conversion unit 640 by using the noise characteristics stored in the noise table 710. The inverse FFT section 670 inverse FFTs the noise-removed power spectrum. The normalization unit 680 normalizes the output of the inverse FFT unit 670 using the pitch information received from the pitch extraction unit, sets the maximum value of the inverse FFT result to "1", and normalizes the waveform representing the waveform in the pitch. The sequence 682a is output as the column normalized waveform sequence 682. In addition, the output level information of the reverse FFT unit 670 is output as level information 681.

【0032】

By the way, the noise removal by the noise removing unit 660 is performed as follows, for example. That is, the noise removing unit 660 creates a noise mask table that stores weights of 0.0 to 1.0 for each frequency according to the stored contents of the noise table 710, and corresponds to the power spectrum sequence at the time of sound. Multiply the weights. This weight is weighted from 0.0 to 1.0 in order so that the frequency having a large absolute value of the power-spelltle (noise power-spectrum) at the time of silence stored in the noise table 710 becomes smaller. In other words, by multiplying the frequency at which noise is prominent by a value of 1.0 or less, the corresponding power spectrum value of the power spectrum at the time of sound is shifted in the direction of decreasing from the original value, and noise is generated. -By giving a weight of 1.0 to the frequency at which the spectrum does not appear, the power spectrum value of that frequency remains as it is. As a result, a noise-free power spectrum sequence can be obtained.

【0033】

Hereinafter, a second embodiment of the present invention will be described.

【0034】

This second embodiment differs from the first embodiment only in the configurations of the sound / silence determination unit 500 and the analysis unit 600.

【0035】

As shown in FIG. 3, the sound / silence determination unit 500 according to the second embodiment is an FFT data setting unit 610 that sets data for FFT processing of frame data and a means for FFT the frame data. The FFT unit 620, the power spectrum conversion unit 630 that obtains the sum of squares of the obtained complex numbers, and the voice power determination that takes the sum of the frame data from the above frame data, compares it with the threshold value, and outputs the sound / silence judge 721. Pitch information is extracted from the unit 510, the sound / silence determination unit 720 that determines the sound noise / silence judge, the FFT unit 620 for FFT processing the power spectrum, the logarithm of the power spectrum, and the cepstrum. However, when there is no sound, the pitch period is not fixed constantly (Digital signal processing Sadayoshi Furui Cepstrum P57 ~ P59 4.9 Pitch extraction) to determine the sound state and output the sound / silence judge. It has a pitch extraction unit 650.

【0036】

By the way, the voice power determination unit 510 determines sound / silence from the power of each frame data, similarly to the voice power determination unit 510 according to the first embodiment. However, according to the above-mentioned determination method, since the sound / silence is determined only by the power, the silence state may be erroneously determined as the sound. Therefore, in the second embodiment, when the voice power determination unit 510 determines that there is sound, the sound / silence determination unit 720 activates the FFT unit 730 and the pitch extraction unit 650, and further uses the pitch cycle. Makes a sound / silence judgment.

【0037】

That is, when the voice power determination unit 510 determines that there is sound, the FFT unit 730 performs FFT processing on the power spectrum. The pitch extraction unit 650 extracts the pitch information from the logarithm of the power spectrum and the cepstrum from this output, and the pitch period is not fixed when there is no sound ("Digital signal processing" Sadayoshi Furui chopsticks P57 to P59 4.9 Pitch extraction. Use (see) to determine the sound / silence state and output the sound / silence judge. The pitch extraction unit 650 determines whether the pitch is sounded or not by converting the pitch information on the time axis into the pitch period on the frequency axis, and if there is a maximum value for each pitch period with respect to the power spectrum, it is sounded. If it does not exist, it may be determined as silence.

【0038】

Next, as shown in FIG. 4, the analysis unit 600 according to the second embodiment includes an axis conversion unit 640 that converts the vertical axis from the power spectrum axis to the amplitude axis, and a noise removal unit 660 that removes noise. Reverse FFT section 670, which is a means of reverse FFT, normalization section 680 to set the maximum value of the reverse FFT result to "1", sound / silence section 920, and noise processing to extract noise characteristics. It has a unit 690, a silence data setting unit 700 for setting data for outputting silence, and a noise table 710. The individual operation of each part is the same as that of the corresponding part of the first embodiment. However, in the second embodiment, the noise removing unit 660, the inverse FFT unit 670, and the normalizing unit 680 constitute a sound processing system. Then, the sound / silence unit 920 causes the sound processing system to perform processing only when the sound / silence determiner output by the pitch extraction unit 650 indicates sound. On the other hand, the processing of the silence processing system composed of the noise processing unit 690 and the silence data setting unit 700 is performed by the sound / silence unit 920 and the sound / silence unit 500 of the sound / silence determination unit 500. This is done when at least one of them determines silence.

【0039】

As described above, according to the present embodiment, by constantly extracting the constantly changing ambient noise and removing the noise feature, it is possible to perform accurate noise removal in any situation.

【0040】

By the way, the processing performed by each of the sound / silence determination unit 500 and the analysis unit 600 according to the first embodiment and the second embodiment can be realized as a program operating on the processor. In this case, when there is no sound, only the silence processing system processing needs to be performed, and when there is sound, only the sound processing system processing needs to be performed, and the silence processing system preprocessing and the sound processing system preprocessing are common. Therefore, the processing load of the processor is smaller than that of the conventional technique of performing noise removal as a preprocessing of voice analysis, and it can be realized as a program running on a single processor.

【0041】

In the above embodiment, the application to the communication system has been described as an example, but the voice information analyzer according to the first and second embodiments also uses the analysis result of the analysis unit 600 to perform voice. It can be applied to various devices such as devices that perform processing such as recognition.

【0042】

[Effect of the invention]

As described above, according to the present invention, it is possible to provide a voice analyzer capable of efficiently removing noise whose characteristics change dynamically from a voice signal to be voice-analyzed.

[Simple explanation of drawings]

[Figure 1]

It is a block diagram which shows the structure of the sound / silence determination part which concerns on 1st Embodiment of this invention.

[Figure 2]

It is a block diagram which shows the structure of the analysis part which concerns on 1st Example of this invention.

[Fig. 3]

It is a block diagram which shows the structure of the sound / silence determination part which concerns on 2nd Example of this invention.

[Fig. 4]

It is a block diagram which shows the structure of the analysis part which concerns on 2nd Example of this invention.

[Fig. 5]

It is a block diagram which shows the structure of the communication system which concerns on embodiment of this invention.

[Explanation of symbols]

100 Voice information analyzer 200 voice input device 300 A / D converter 400 buffer memory 401 frame data 500 Sound / Silence Judgment Unit 511 Sound / Silence Judge 600 Analysis Department 610 FFT data setting unit 620 FFT section 630 Power spectrum converter 631 Power spectrum sequence 640 Axis converter 650 pitch extractor 651 Pitch information 660 Noise remover 670 Reverse FFT section 680 Normalization section 681 Normalized waveform series 682 Level information 690 Noise processing unit 700 Silence frame setting 710 Noise Table 720 Sound / Silence 800 Vector quantization unit 801 vector code 900 transmitter 1100 Vector inverse quantization unit 1200 synthesis section 1300 D / A converter 1400 audio output device