JP2007068847A - Glottal closure region detecting apparatus and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a glottis closing zone detecting apparatus detecting a glottal closure region in a natural phonation state with a simple constitution. <P>SOLUTION: A changeover device 202 directly receives an analog voice signal from a microphone 132 and receives a signal after the voice signal from the microphone 132 passes through a passing band variable BPF (Band-Pass Filter) 200. The output of the changeover device 202 is converted into a digital signal by an A/D converter 204 and then stored in a buffer memory section 206. CPU 120 detects a forth formant frequency region based on the frequency spectrum from a frequency analysis part 208 and controls the passing band of the BPF 200 to pass through the fourth formant frequency region based on the detection. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a glottal closing segment detecting device and a glottal closing segment detecting method capable of detecting a glottal closing segment when a voice is uttered from a voice signal.

  Human vocal cord folds vibrate 100 times or more, sometimes 1,000 times per second during utterance. For this reason, the state of each vibration cannot be directly seen with the naked eye.

  Therefore, in order to visualize the vibration of the vocal cords, there are a direct method for visualizing the larynx image by application of the device and an indirect method for detecting and recording only the opening and closing movement of the glottis. Can be broadly classified. Direct observation methods include laryngeal high-speed movies, laryngeal stroboscopic copying, photochromography, and semiconductor imaging device methods, and indirect observation methods include photoelectric glography, electro-glottogram (EGG), ultra There is sonic grography.

  Among them, EGG is a method of detecting and recording a change in electrical impedance due to opening and closing of the glottis by placing electrodes on the skin surfaces outside the left and right thyroid cartilage plates and passing a high-frequency current.

  According to Non-Patent Document 1, as shown in FIG. 7, the movement of vocal cord mucosa during utterance is not a simple left-right opening / closing movement, but a three-dimensional movement accompanied by a vertical wave. In FIG. 7, 1 to 3 indicate the glottal opening period, 3 to 7 indicate the closing period, and 7 to 10 indicate the closing period.

  However, when considering vocalization at the larynx, the change in glottal area on a plane perpendicular to the airflow is the most problematic, so glottal area waveform (glottal area is displayed as a function of time) when observing glottal vibration. Is the most important issue.

  FIG. 8 is a diagram showing such a glottal area waveform. In the glottal area waveform, for each vibration cycle, the above-described three phases of the large period, the small period, and the closed period are distinguished. The time required for one vibration is called a basic period. The number of vibrations per unit time is called the fundamental frequency. The fundamental period of speech coincides with the fundamental period of vocal cord vibration. Therefore, the fundamental frequency of speech is equal to the fundamental frequency of vocal cord vibration.

  On the other hand, in voice transmission / recognition, it is extremely important to accurately estimate vocal tract transmission characteristics, and a linear prediction method has been conventionally used as one of the estimation methods. However, in order to obtain accurate vocal tract transfer characteristics using normal linear prediction methods, the excitation source must be a single impulse or white noise. However, in reality, this assumption does not hold, and the influence of the excitation source occurs on the formant frequency estimation.

In order to reduce the influence of such an excitation source, the analysis window length is shortened to 1 pitch period or less, and only the glottal closing (closed) period, that is, the free vibration period is estimated, and this is analyzed (for example, In the sample selection linear prediction method that selects speech sample points that match the linear prediction model by referring to residual information (see Non-Patent Document 2), the overall characteristics of the prediction error are considered in the sample selection linear prediction method. And performing this process in two stages to propose a “two-stage sample selection linear prediction method” or the like that removes speech samples in the glottal opening period from non-predicted samples (see, for example, Non-Patent Document 3). .
Edited by the Japan Spoken Language Society, Bio-Dental Publishing Co., Ltd. 97-p. 99, 1994 K. Steiglitz and B. Dickinson: “The use of time-domain selection for improved linear prediction”, IEEE Tran. Acoust., Speech & Signal process, ASSp-25, pp. 34-39 (1977) Yoshiaki Miyoshi, Kazuharu Yamato, Masuzou Yanagita, Toru Kakusho: “Analysis of high pitch speech using two-stage sample selection linear prediction method”, IEICE Transactions A Vol. J70-A No.8 pp.1146- 1156 August 1987

  However, in the former, it is generally difficult to accurately estimate the glottal closure interval of natural speech, and in the latter case, those whose absolute value of the residual is greater than or equal to the threshold value are excluded from the predicted sample. What is being processed does not compare with the observation of glottal state.

  In addition, if it can easily detect the glottal closure interval, it can be used as a clear voicing index in voice training, etc. Although it can be expected that it can be utilized, the above-mentioned method of observing vocal cord vibration is not suitable for observation during natural utterance, or for the purpose of rehabilitation because it places a physical or mental burden on the subject for measurement. There were problems such as being unsuitable.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to detect a glottal closure interval detection device that can detect a glottal closure interval in a natural utterance state with a simple configuration. And providing a glottal closure interval detection method.

  In order to achieve such an object, according to one aspect of the present invention, there is provided a glottal closed section detecting device that selects an audio signal in a frequency band corresponding to laryngeal cavity resonance from among input audio signals. Band extracting means for extracting automatically, and arithmetic means for determining the glottal closure interval based on the intensity of the extracted voice signal.

  Preferably, the band extracting means includes band-pass type filter means that can change the pass band variably, and the glottal closed section detecting device frequency-analyzes the input voice signal to obtain a frequency band corresponding to laryngeal cavity resonance. In particular, it further comprises a passband setting means for setting as a passband of the bandpass filter means, the computing means according to when the intensity of the extracted audio signal exceeds a set threshold value, The corresponding voice signal section is determined as a glottal closed section.

  Preferably, the passband setting means determines the fourth formant as a frequency corresponding to the laryngeal cavity resonance based on the frequency spectrum of the audio signal.

  According to another aspect of the present invention, there is provided a glottal closed section detecting method, wherein a voice of a subject is converted into a voice signal and a voice signal in a frequency band corresponding to laryngeal cavity resonance is selectively extracted; Determining a glottal closure interval based on the intensity of the voice signal that has been generated.

  Preferably, in the extracting step, a frequency analysis corresponding to the laryngeal cavity resonance is performed by analyzing the frequency of the input audio signal, and the frequency band corresponding to the identified laryngeal cavity resonance is band-pass filter means. The step of determining as a passband of the vowel is determined to be a glottal closed section in response to the fact that the output intensity of the bandpass filter means exceeds a set threshold value. Determining.

  According to the glottal closed section detecting device and the glottal closed section detecting method according to the present invention, it is possible to detect the glottal closed section with a simple device configuration without requiring a special device.

  Moreover, according to the glottal closed section detecting device and the glottal closed section detecting method according to the present invention, it is possible to detect the glottal closed section in the natural utterance state of the subject regardless of the content of the utterance.

Embodiments of the present invention will be described below with reference to the drawings.
[Hardware configuration]
FIG. 1 is a block diagram showing an example of a glottal closing segment detection device 100 to which the glottal closing segment detection method of the present invention is applied.

  Referring to FIG. 1, glottal closing section detecting device 100 is basically configured by providing a voice processing interface in a personal computer.

  That is, the glottal closed section detecting device 100 reads / writes information to / from an optical disc drive 108 for reading information on an optical disc such as a CD-ROM (Compact Disc Read-Only Memory) 118 and a flexible disc (FD) 116. A computer main body 102 provided with an FD drive 106, a monitor 104 as a display device connected to the computer main body 102, a keyboard 110 and a mouse 112 as input devices also connected to the computer main body 102, and voice input A microphone 132 as a device and a speaker 134 as an audio output device are included.

  In addition to the optical disk drive 108 and the FD drive 106, the computer main body 102 includes a CPU (Central Processing Unit) 120 that is an arithmetic processing unit connected to the bus BS, a ROM (Read Only Memory), and a RAM (Random Access Memory). ), A direct access memory device, for example, a hard disk 124, and a voice processing interface unit 128 for exchanging data with the microphone 132 or the speaker 134.

  The CD-ROM 118 may be another medium, such as a DVD-ROM (Digital Versatile Disc) or a memory card, as long as it can record information such as a program installed in the computer main body. In this case, the computer main body 102 is provided with a drive device that can read these media.

  The main part of the glottal closing section detecting device of the present invention is constituted by computer hardware and software for controlling the glottal closing section detecting device executed by the CPU 120. Generally, such software is stored and distributed in a storage medium such as a CD-ROM 118 or FD 116, read from the storage medium by the CD-ROM drive 108 or FD drive 106, and temporarily stored in the hard disk 124. Alternatively, when the device is connected to the network 310, it is temporarily copied from the server on the network to the hard disk 124. Then, the data is further read from the hard disk 124 to the RAM in the memory 122 and executed by the CPU 120. In the case of network connection, the program may be directly loaded into the RAM and executed without being stored in the hard disk 124.

  The computer hardware itself shown in FIG. 1 and its operating principle are general. Therefore, the most essential part of the present invention is software stored in a storage medium such as the FD 116, the CD-ROM 118, and the hard disk 124.

  As a general tendency, various program modules are prepared as a part of a computer operating system, and an application program generally calls a module in a predetermined arrangement and advances the processing when necessary. In such a case, the software itself for realizing the glottal closing section detecting device does not include such a module, and the glottal closing section detecting device is realized only in cooperation with the operating system on the computer. However, as long as a general platform is used, it is not necessary to distribute software including such modules, and the software itself not including these modules and the recording medium storing the software (and the software distributes on the network). Data signal) can be considered to constitute the embodiment.

  FIG. 2 is a functional block diagram for explaining the configuration of the voice processing interface unit 128 shown in FIG. 1 in more detail. In FIG. 2, only the part related to the input processing of the audio signal from the microphone 132 is extracted and shown.

  Referring to FIG. 2, switcher 202 directly receives an analog audio signal from microphone 132, and after the audio signal from microphone 132 passes through a bandpass filter (hereinafter referred to as BPF) 200 having a variable passband. Receive the signal. The CPU 120 controls which passband of the BPF 200 and which signal the switcher 202 selects.

  The output of the switcher 202 is converted into a digital signal by the A / D converter 204 and then stored in the buffer memory unit 206. The frequency analysis unit 208 obtains a frequency spectrum for the audio signal stored in the buffer memory unit 206 and outputs it to the CPU 120.

  The CPU 120 detects a formant (fourth formant) frequency region corresponding to the laryngeal cavity resonance based on the frequency spectrum, and based on this, detects the frequency region of the BPF 200 so as to pass through the region of the fourth formant. Control the passband. Further, the result of the processing by the CPU 120 is displayed on the display device 104, for example.

  In FIG. 2, the BPF 200 has been described as using an analog band variable filter. However, it is also possible to use a digital band-variable filter by arranging the BPF 200 and the switching unit 202 at the subsequent stage of the A / D converter 204. Alternatively, the audio signal from the microphone 132 is converted into a digital signal by the A / D converter 204 and directly stored in the buffer memory unit 206, and the CPU 120 performs arithmetic processing on the audio signal data in the buffer memory 206. It is good also as performing a digital filter process by performing.

  Further, the buffer memory unit 206 is not necessarily provided in the audio processing interface unit 128. For example, the memory 122 or the hard disk 124 may be used as the buffer memory. Further, the frequency analysis unit 208 is not necessarily provided in the audio processing interface unit 128, and for example, equivalent processing can be performed by arithmetic processing such as Fourier transform of the CPU 120.

  FIG. 3 is a diagram showing the result of the effect of the glottal opening area Ag on the vocal tract transmission characteristics obtained by the equivalent circuit model.

In FIG. 3, the glottal opening area Ag 0.0 cm ² (glottal closure), 0.1 cm ^2, is varied in three steps of 0.2 cm ^2. It can be seen from this figure that the 3.1 kHz formant (fourth peak from the low frequency side: the fourth formant) disappears due to the opening of the glottis. This formant is consistent with the formant produced by the laryngeal cavity (laryngeal cavity resonance).

  Therefore, it is predicted that the laryngeal cavity resonance appears in the closed section of the glottis and disappears in the open section. Therefore, it is considered that the glottal closure section can be extracted from the voice by detecting the intra-periodic variation of the laryngeal cavity resonance.

  FIG. 4 is a flowchart for explaining the operation of the glottal closing section detecting device 100 shown in FIGS. 1 and 2.

  Referring to FIG. 4, first, under the control of CPU 120, switcher 202 provides a signal received directly from microphone 132 to A / D converter 204, and a digital audio signal is stored in buffer memory unit 206. The The frequency analysis unit 208 performs frequency analysis on the data in the buffer memory unit 206 (S100).

  The CPU 120 analyzes the frequency spectrum obtained as a result of the frequency analysis, thereby specifying the frequency region of the fourth formant of the subject (S102).

  Subsequently, the CPU 120 adjusts the passband of the BPF 200 so as to pass the fourth formant audio signal (S104). Although not particularly limited, for example, if the peak position of the fourth formant is known, adjustment may be made so that a signal of a band corresponding to a predetermined frequency is passed above and below the frequency.

  After adjusting the pass band of the BPF 200 in this way, the CPU 120 controls the switcher 202 to adjust so that the signal that has passed through the BPF 200 is stored in the buffer memory unit 206. Thereafter, for the same input condition for the same subject, the closed glottal section is detected according to the adjusted signal intensity from the BPF 200 (S106). That is, since the signal strength from the BPF 200 becomes large in the glottal closed section, the threshold value is set and the CPU 120 can determine that the section where the signal strength exceeds the threshold value is the glottal closed section. . Although not particularly limited, such a threshold value may be set manually by the user looking at the measurement result output to the display device 104, or depending on the intensity of the signal that the CPU 120 has passed through the BPF 200, For example, you may set so that it may become the predetermined ratio of the absolute value of the highest intensity | strength.

(Experimental result)
FIG. 5 and FIG. 6 are diagrams showing the results of recording Japanese vowels / a / and / i / that were uttered continuously by one male and one female in a sitting position in an anechoic room. FIG. 5 shows measurement results for men, and FIG. 6 shows measurement results for women.

  In the experiments shown in FIG. 5 and FIG. 6, EGG signals were recorded simultaneously with voice. The direct current component was removed from the EGG signal by a high-pass filter with a cutoff of 1.6 Hz. These signals were recorded at a sampling frequency of 48 kHz and a quantization of 16 bits. Since there is a time difference corresponding to the distance from the glottis to the microphone between the voice and the EGG signal, the EGG signal is shifted by this time difference.

  As the BPF 200, an FIR (Finite Impulse Response) type band-pass filter was created by applying a window function to the Fourier series of ideal filter characteristics. A 101-point Hamming window was used as the window function.

  From the spectrogram of the voice data, the passband of the band pass filter of the male speaker was determined from 2.8 kHz to 3.8 kHz, and the passband of the female speaker was determined from 3.8 kHz to 4.8 kHz. The bandpass filter was applied to the audio data, and the output was compared with the EGG signal.

  FIGS. 5 and 6 also show the 30 msec speech waveform of the vowel, the corresponding EGG signal, the output of the bandpass filter (fourth formant signal) and the second formant signal for comparison. Since the EGG signal is proportional to the contact area of the vocal cords, a section with a large value is a glottal closure section.

  5 and 6 that the amplitude of the output of the bandpass filter becomes relatively large in the glottal closed period. Similar to the simulation results shown in FIG. 3, this result shows that the laryngeal cavity resonance (fourth formant) appears in the glottal closed section of the pitch period even in real speech, and this resonance disappears in the glottal open section. ing. Within one cycle of vocal cord vibration, the glottis closes rapidly and opens slowly. The bandpass filter output also corresponds to this, and the amplitude increases rapidly at the start of glottal closure and then gradually attenuates.

  Accordingly, the on / off change can be clearly detected in the bandpass filter output, that is, the envelope of the signal corresponding to the fourth formant, and the glottal closure interval can be determined by threshold processing. On the other hand, the second formant is gently attenuated, and a clear on / off change cannot be detected.

  As described above, it is possible to detect the glottal closure section by utilizing the fact that the laryngeal cavity resonance pattern of the vowel varies within one pitch period. The method for detecting the glottal closure interval of the present invention can be applied to the back tongue vowel, and can record glottal opening and closing in a more natural voicing state. This method can be basically realized by a computer having an audio input / output function if a microphone and a band-pass filter are used. Furthermore, unlike the other formants, the laryngeal cavity resonance appears in a substantially constant frequency band regardless of the vowels, so that any vowel can be used once the passband of the bandpass filter is determined.

  In voice training or the like, the detected glottal closure interval can be used as a clear utterance index. Alternatively, in speech and auditory therapy, it can be used for diagnosis of glottic insufficiency speech and rehabilitation support.

  In the above description, only the detection of the glottal closed section has been described. However, in general, the speech processing technology is based on the premise that the glottis are closed when speaking. Therefore, it is necessary to extract only from the glottal closed section when extracting the voice feature value. An application is also possible in which the glottal closed section is detected by using the glottal closed section detection method of the present invention, and the feature amount is extracted therefrom.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

It is a block diagram which shows an example of the glottal closed area detection apparatus 100 with which the glottal closed area detection method of this invention is applied. 3 is a functional block diagram for explaining the configuration of a voice processing interface unit 128 in more detail. FIG. It is a figure which shows the result of the influence of the glottal opening area Ag in the vocal tract transmission characteristic calculated | required with the equivalent circuit model. 4 is a flowchart for explaining the operation of the glottal closure section detection device 100. It is a figure which shows the result of having recorded the Japanese vowel / a / and / i / which the male test subject continuously uttered in the sitting position in the anechoic room. It is a figure which shows the result of having recorded the Japanese vowel / a / and / i / which the female test subject uttered continuously in the sitting position in the anechoic room. It is a conceptual diagram which shows the movement of the vocal cord mucosa during utterance. It is a figure which shows a glottal area waveform.

Explanation of symbols

  100 Glottal Closure Section Detection Device, 102 Computer Main Body, 104 Display Device, 106 FD Drive, 108 Optical Disk Drive, 110 Keyboard, 112 Mouse, 116 Flexible Disk, 118 CD-ROM, 120 CPU, 122 Memory, 124 Hard Disk, 128 Audio Processing Interface unit, 132 microphone, 134 speaker, 200 BPF, 202 switching unit, 204 A / D converter, 206 buffer memory unit.

Claims (5)

Band extraction means for selectively extracting an audio signal in a frequency band corresponding to laryngeal cavity resonance from the input audio signal;
A glottal closure segment detection device comprising: a calculating means for determining a glottal closure segment based on the intensity of the extracted voice signal. The band extracting means includes band pass filter means that can change the pass band variably,
A frequency analysis of the input audio signal to identify a frequency band corresponding to the laryngeal cavity resonance, further comprising passband setting means for setting as the passband of the bandpass filter means;
2. The glottal closing section according to claim 1, wherein the computing means determines that the corresponding speech signal section is the glottal closing section when the intensity of the extracted voice signal exceeds a set threshold value. 3. Detection device.   The glottal closure segment detection device according to claim 2, wherein the passband setting means determines the fourth formant as a frequency corresponding to the laryngeal cavity resonance based on a frequency spectrum of the audio signal. Converting the subject's voice into a voice signal and selectively extracting a voice signal in a frequency band corresponding to the laryngeal cavity resonance;
A glottal closure interval detection method comprising: determining a glottal closure interval based on the intensity of the extracted speech signal. The extracting step includes:
Analyzing the frequency of the input audio signal to identify a frequency band corresponding to the laryngeal cavity resonance;
Setting a frequency band corresponding to the identified laryngeal cavity resonance as a pass band of the band-pass filter means,
The step of determining includes
5. The glottal closed section according to claim 4, comprising the step of determining a corresponding voice signal section as the glottal closed section in response to an output intensity of the band-pass filter means exceeding a set threshold value. Detection method.

Images