JP2004341340A - Speaker recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To precisely recognize a speaker and to easily calculate the distance to a speaker model. <P>SOLUTION: In a frame trimming means 21, speech waveform data are successively trimmed to the frames with a frame width L at a shift interval T for outputting to a feature vector generating means 22. The presence or absence of pitch in a frame trimmed by the frame cutting means is detected by a pitch detection means 23. The frame trimming means changes the shift interval T and the frame width L according to the pitch detection result. More specifically, the shift interval T and the frame width L between sounds, where a sound frequency exists to a silent section, are changed to be shorter. A feature vector generating means converts the sound waveform data of each frame to a feature vector for outputting to a next-stage distance calculation means 24. The distance calculation means calculates the distance between a feature vector series and a speaker model stored in a speaker model storage means 25 for outputting to a next-stage recognition means 26. The recognition means compares the distance data from the distance calculation means and a preset threshold to recognize the speaker. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a speaker recognition device that recognizes a speaker by inputting voice.
[0002]
[Prior art]
In general, in speaker recognition, an input voice is cut out at a cyst interval of about 10 msec into a frame having a length of about 30 msec, and various personality features obtained for each frame are extracted as a feature vector sequence over the entire voiced sound section. The extracted feature vector sequence is evaluated by a comparison method such as the VQ distortion method or the HMM method to perform speaker recognition.
In recent years, it has been known to use not only a voiced sound section but also an unvoiced sound section in recognizing the personality of voice (for example, see Non-Patent Document 1).
[0003]
[Non-patent document 1]
By Tomoko Matsui and Sadahiro Furui, IEICE Technical Report SP90-26, "Text-independent speaker recognition using sound source / vocal tract features", IEICE, June 1990, pp. 55-62
[0004]
[Problems to be solved by the invention]
By the way, it is known to perform weighting based on the feature amount when performing speaker recognition. However, when such weighting is applied to one that uses both a voiced sound section and an unvoiced sound section, a frame to be divided is divided. While cutting out frames with a fixed length, weighting is applied not only to voiced sections but also to unvoiced sections, but voiced sections are weighted twice as much as unvoiced sections, for example. become. However, if the frame length from which both the unvoiced sound section and the voiced sound section are cut out is fixed, the analysis in the voiced sound section becomes coarse. Then, the coarse part is weighted to cope with the problem, but such processing does not allow sufficient speaker recognition. Further, when calculating the distance between the analysis result and a preset speaker model, a weighting process must be performed each time, and the process is troublesome.
The present invention provides a speaker recognition device that can perform speaker recognition with high accuracy and that can easily calculate a distance from a speaker model.
[0005]
[Means for Solving the Problems]
The present invention relates to a voice input unit for inputting voice, for example, a microphone or the like, and a frame cutout for cutting out voice waveform data input by the voice input unit into a frame having a set frame width at a set shift interval. Means, a feature vector generating means for converting the speech waveform data of the frame extracted by the frame extracting means into a feature vector, a pitch detecting means for detecting the presence or absence of a vocal cord frequency in the frame extracted by the frame extracting means, A speaker model storing means for storing a model; and a recognition means for performing speaker recognition based on a distance between a feature vector generated from the feature vector generating means and a speaker model stored in the speaker model storing means, and frame extraction. The means shall be set so that the shift interval becomes short when the pitch detecting means detects a vocal cord frequency. A.
[0006]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of the entire apparatus. 1 is a CPU (central processing unit) constituting a control unit main body, and 2 is a ROM (read only memory) storing program data for controlling each unit by the CPU 1. Reference numeral 3 denotes a RAM (random access memory) provided with a memory used by the CPU 1 for data processing and various buffer memories.
[0007]
4 is an I / O port, 5 is a hard disk drive provided with speaker model storage means for storing a number of speaker models in advance, 6 is a keyboard controller for taking in key signals from a keyboard 7, and 8 is a display 9 This is a display controller that displays information such as text and images.
The CPU 1, ROM 2, RAM 3, I / O port 4, hard disk 5, and controllers 6, 8 are electrically connected to each other via a bus line 10.
[0008]
Further, a microphone 11 for taking in sound is provided, and this microphone 11 converts the taken sound into an electric analog signal and outputs it, and supplies the signal to a low-pass filter 12 in the next stage. The low-pass filter 12 cuts a frequency equal to or higher than a predetermined frequency from the input analog signal, outputs the cut signal, and inputs the cut signal to an A / D converter 13 at the next stage. The A / D converter 13 converts the input analog signal into a digital signal at a predetermined sampling frequency and a predetermined number of quantization bits, and supplies the digital signal to the I / O port 4 as audio waveform data. The microphone 11, the low-pass filter 12, and the A / D converter 13 constitute an audio input unit.
[0009]
FIG. 2 is a functional block composed of a complex of the CPU 1, the ROM 2, the RAM 3, and the hard disk device 5. The frame extracting unit 21 converts the audio waveform data input from the I / O port 4 into a set shift. The frames are sequentially cut into frames having a frame width L set at the interval T, and are supplied to the feature vector generating means 22 in the next stage.
[0010]
The pitch detecting means 23 detects the presence or absence of a voice frequency, that is, the presence or absence of a pitch in the frame extracted from the frame extracting means 21, and outputs the detection result to the frame extracting means 21. The frame cutout means 21 changes the shift interval T and the frame width L set according to the detection result by the pitch detection means 23. That is, the shift interval T and the frame width L of the voiced section in which the voice frequency exists are changed to be shorter than the unvoiced section in which the voice frequency does not exist.
[0011]
The feature vector generation unit 22 converts the audio waveform data of each frame that is sequentially input into a feature vector such as a cepstrum coefficient and outputs the feature vector to the next-stage distance calculation unit 24 as a feature vector sequence. . The distance calculating means 24 calculates the distance between the input feature vector sequence and a speaker model such as a codebook stored in the speaker model storing means 25 provided on the hard disk 5, and recognizes the next stage. Output to the means 26.
The recognition means 26 performs speaker recognition by comparing the distance data from the distance calculation means 24 with a preset threshold value, and displays the result on, for example, the display 9.
[0012]
FIG. 3 is a flowchart showing processing by the functional blocks in FIG. 2. First, in S1, the frame cutout means 21 sets the shift interval T and the frame width L to T ₀ and L ₀ corresponding to the unvoiced sound section. Subsequently, at S2, the frame cutout section 21 a speech waveform data temporarily stored in the buffer memory RAM3 crowded taken from the I / O port 4, the frame width L ₀ which is set in the shift interval T ₀ that has been set Cut out sequentially into frames.
[0013]
Analysis processing is performed on the audio waveform data of the cut-out frame in S3, and pitch detection by the pitch detection means 23 is performed in S4. If pitch detection is not performed, that is, if no voice waveform is detected, the end of the voice is checked in S5. If not, the process returns to S1 and repeats the processing.
[0014]
The setting, the pitch detection unit 23 performs pitch detection, in S6, in extracts pitch frequency _{f P,} S7, the frame width L sets the shift interval T to _{T 1} to F _{(f P)} I do. In this case, T ₁ <T ₀ and F (f _P ) <L ₀ . F (f _P) is a function for obtaining the frame width L from pitch frequency f _P, the frame width L when performing pitch detection indicates that changes in accordance with the pitch frequency f _P.
Set the shift interval T to T _1, setting the frame width L to F (f _P), the process is repeated by performing the extraction of the speech waveform returns to S2. When the end of the voice is checked in S5, a recognition process is performed based on the distance for each frame stored in the RAM 3 in advance in S8, and a recognition result is output.
[0015]
The analysis processing in S3 is the processing shown in FIG. That is, in S31, feature vector generation processing by the feature vector generation means 22 is performed, in S32, distance calculation processing by the distance calculation means 24 is performed, and in S33, one frame calculated by the distance calculation means 24 is processed. The distance for each is stored in the RAM 3.
[0016]
In such a configuration, when a sound is input from the microphone 11, the sound is converted into an electric analog signal, and a frequency higher than a predetermined frequency is cut by the low-pass filter 12. In this case, a frequency equal to or more than 1/2 of the sampling frequency in the A / D converter 13 is cut. For example, if the sampling frequency is 12 kHz, frequencies above 6 kHz are cut off. The analog signal that has passed through the low-pass filter 12 is sampled at a sampling frequency of 12 kHz in an A / D converter 13 and converted into a digital signal.
[0017]
The audio waveform data converted into a digital signal by the frame extracting means 21 is sequentially extracted into frames having a frame width L at a shift interval T. Then, the presence or absence of a pitch in the frame cut out by the frame cutout means 21 is detected by the pitch detection means 23, and the frame cutout means 21 sets the shift interval T to T ₀ between a state where the pitch is not detected and a state where the pitch is detected. To T ₁ and the frame width L from L ₀ to F (f _P ). For example, the shift interval T is converted from 6 msec to 3 msec, and the frame width L is converted from 24 msec to about 12 msec. In addition, the reason why it is set to about 12 msec is that it changes depending on the pitch frequency.
[0018]
That is, as shown in FIG. 5, a frame is cut out with a frame width of 24 msec while shifting at a shift interval of 6 msec in a section where pitch detection is not performed, as shown as frames 101 to 105, with respect to the input speech waveform 100, When the pitch detection is performed, the frame width of 24 msec is converted to about 12 msec as shown as a frame 106, and thereafter, the frame having the frame width of about 12 msec is cut out while sequentially shifting at a shift interval of 3 msec.
[0019]
Then, the speech waveform data of the sequentially cut out frame is converted into a feature vector sequence by the feature vector generation means 22, and the feature vector series and the speaker model stored in the speaker model storage means 25 by the distance calculation means 24. Is calculated, and the distance data calculated by the recognition unit 26 is compared with a threshold. Thus, speaker recognition is performed, and the result is displayed on the display 9.
[0020]
As described above, when the input speech waveform is cut into frames, the shift interval T and the frame width L are increased in a section where the pitch is not detected, and the shift interval T and the frame width L are decreased after the pitch is detected. The number of frames per unit time is relatively larger in voiced sound sections than in unvoiced sound sections. That is, a frame is cut out more finely in the voiced sound section than in the unvoiced sound section. Therefore, the specific gravity in the speaker recognition is increased, and highly accurate speaker recognition can be performed as in the case of weighting.
[0021]
In addition, since the shift interval T and the frame width L for the voiced sound section are reduced and the number of frames to be cut out is increased to improve the accuracy, more accurate pitch detection can be performed as compared with the case of weighting. There is no need to perform troublesome weighting, and the distance to the speaker model can be easily calculated.
[0022]
In addition, since the setting of the frame width L in the voiced sound section is performed using the pitch frequency f _P , that is, the function F (f _P ) that changes according to the frequency of the vocal cord frequency, the frame width L in the voiced sound section is set. Appropriate settings can be made according to the vocal cord frequency, and the pitch detection accuracy can be maintained regardless of the vocal cord frequency.
[0023]
【The invention's effect】
As described in detail above, according to the present invention, it is possible to provide a speaker recognition device that can perform speaker recognition with high accuracy and can easily calculate a distance from a speaker model.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an entire apparatus according to an embodiment of the present invention.
FIG. 2 is a block diagram functionally showing a main part configuration in the embodiment.
FIG. 3 is a flowchart showing main part processing in the embodiment.
FIG. 4 is a flowchart showing the contents of an analysis process in FIG. 3;
FIG. 5 is an exemplary view for explaining frame extraction for an input audio waveform according to the embodiment;
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... CPU, 11 ... Microphone, 13 ... A / D conversion part, 21 ... Frame extraction means, 22 ... Feature vector generation means, 23 ... Pitch detection means, 24 ... Distance calculation means, 25 ... Speaker model storage means, 26 ... recognition means.

Claims (3)

Voice input means for inputting voice, frame cutout means for cutting out the voice waveform data input by the voice input means into frames of a set frame width at a set shift interval, and frame cutout means cut out by the frame cutout means Feature vector generating means for converting the speech waveform data of the frame into a feature vector, pitch detecting means for detecting the presence or absence of a vocal cord frequency in the frame cut out by the frame cutout means, and speaker model storage means for storing a speaker model in advance And recognition means for performing speaker recognition from the distance between the feature vector generated from the feature vector generation means and the speaker model stored in the speaker model storage means,
The speaker recognizing device, wherein the frame cutout means sets the shift interval to be short when the pitch detection means detects a vocal cord frequency. 2. The speaker recognition apparatus according to claim 1, wherein the frame extracting means sets the shift interval to be short and sets the frame width to be short when the pitch detecting means detects a vocal cord frequency. 3. The speaker recognition apparatus according to claim 2, wherein the frame extracting means changes a frame width to be set according to a vocal cord frequency.

Images