JPS6329759B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a plosive sound detection device for detecting plosive sounds.

Previously, only plosive sounds could be reliably detected from human speech.
There is no known method for simple and real-time detection.

An object of the present invention is to provide a plosive sound detection device that reliably, simply, and in real time detects only plosive sounds based on the flow velocity of oral airflow during speech and voice information.

Experiments have shown that the velocity waveform of oral airflow during speech is specific to plosive sounds, and that plosive sounds can be detected effectively by performing simple signal processing on this velocity waveform. . However, it has become clear that this method sometimes detects sounds other than plosives, often voiceless fricatives, as plosives, depending on the way the utterance is made.

The present invention detects plosive candidates from oral airflow information, extracts fricatives from voice information, and removes fricatives from the plosive candidates.
This method correctly detects only plosive sounds.

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of a plosive sound detection device according to the present invention. In the figure, 1 is a flow velocity detection means for detecting the oral air flow velocity, 2 is a plosive sound detection means for detecting plosive candidates based on the output signal of the flow velocity detection means 1, and 3 is a microphone for detecting sound.
4 is a fricative sound extracting means for extracting a fricative sound from the output signal of the microphone 3; and 5 is a determining means for suppressing the output signal of the plosive sound detecting means 2 by the output signal of the fricative sound extracting means 4. The flow velocity detection means 1 measures the velocity of the airflow in front of the mouth caused by speech and outputs an electrical signal, and is composed of, for example, a hot wire type current meter. The plosive sound detection means 2 processes the output signal of the flow rate detection means 1, ie, the so-called oral airflow waveform, to detect plosive sound candidates, and is composed of, for example, a differentiating circuit made up of resistors and capacitors. The microphone 3 detects voice, and is, for example, a low-noise close-talk type microphone. The fricative sound extraction means 4 extracts fricative sounds from the output signal of the microphone 3, which is a so-called audio signal, and is composed of, for example, a low-pass filter circuit, a high-pass filter circuit, and a division circuit, and each filter circuit separates the sound. A division circuit divides the output of the high-pass filter circuit by the output of the low-pass filter circuit, and fricative sounds are extracted depending on the magnitude of the output. The determining means 5 suppresses the output signal of the plosive sound detecting means 2 using the output signal of the fricative sound extracting means 4, and is composed of, for example, a NAND circuit or a variable gain amplifier circuit.

Figures 2 a to e and Figures 3 a to e show signal waveforms of each part of the plosive sound detection device shown in Figure 1. Figure 2 shows a plosive sound, and Figure 3 shows a voiceless fricative sound. If you do it it shows it. Waveforms A and B represent representative examples of audio signals of plosives and voiceless fricatives detected by the microphone 3 during speech, and waveforms C and D represent oral airflow waveforms detected by the flow velocity detection means 1 at that time. . Waveforms E and F are plosive sound detection means 2
This shows waveforms in which candidates for plosive sounds were detected from the oral airflow waveforms c and d. Waveforms ① and ① indicate signals obtained by extracting fricatives from audio signals ① and ② by the fricative sound extracting means 4, that is, fricative sound signal waveforms, and waveforms ① and ① indicate output signals of the determining means 5.

In the above configuration, plosive sounds such as /
When the sound /Pa/ containing P/ is uttered, the flow velocity detection means 1 detects the oral airflow waveform C, and the microphone 3 detects the audio signal A. The plosive sound detection means 2 detects a differential signal E, which is a candidate for a plosive sound, by differentiating the oral airflow waveform C. On the other hand, the fricative sound extracting means 4 extracts a fricative sound from the audio signal A. There are few high frequency components in the audio signal A, so there is almost no output from the high pass filter circuit. Therefore, as shown in waveform G, there is no output from the fricative sound extraction means 4. Therefore, the output of the determining means 5, which suppresses the differential signal E to a small value when the fricative signal G is large, generates a plosive pulse signal R representing a plosive sound since there is no fricative signal G.

Next, for example, a sound containing the unvoiced fricative /S/ /
When Sa/ is uttered with particular emphasis on /S/, the output signal of the flow velocity detection means 1 detects an abnormally large oral airflow waveform D. This oral airflow waveform D may be outputted by the plosive sound detection means 2 as a differential signal close to the differential signal E of the plosive sound. Therefore, the voiceless fricative /S/ is a candidate for a plosive. On the other hand, the fricative sound extracting means 4 extracts a fricative sound from the audio signal B. There are many high frequency components in the audio signal B, and therefore the output from the high pass filter circuit is large. For this reason,
The output of the fricative sound extraction means 4 generates a fricative sound signal as shown in waveform H. Therefore, the output of the determining means 5, which suppresses the differential signal so as to be small when the fricative signal Q is large, does not generate a plosive pulse signal representing a plosive as shown in waveform N.

As described above, according to this embodiment, regardless of the manner of utterance, voiceless fricatives are not detected as plosives, and only plosives can be correctly detected. In addition, without using complicated signal processing means, a current meter, a simple differentiation circuit, a microphone, two filter circuits, and
It can be configured with NAND circuits, etc. Furthermore, it is clear that detection can be performed in real time in principle.

As explained above, the present invention detects plosive candidates from oral airflow information, and on the other hand, extracts fricatives from voice information and removes fricatives from the plosive candidates. Can be detected correctly.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing one embodiment of the plosive sound detection device according to the present invention, and Figs. 2 a to e and 3 a to e show the various parts of the device when a plosive sound and a voiceless fricative sound are uttered, respectively. FIG. 3 is a diagram showing signal waveforms. 1...Flow velocity detection means, 2...Plosive sound detection means,
3...Microphone, 4...Fricative sound extraction means, 5...Determination means.

Claims (1)

[Claims] 1. A flow velocity detection means for detecting oral airflow, a plosive detection means for processing the output signal of the flow velocity detection means and detecting plosive sound candidates, a microphone for detecting sound, and detecting fricative sounds from the output signal of the microphone. A plosive sound detection device comprising: a fricative sound extraction means for extracting a fricative sound; and a means for suppressing an output signal of the plosive sound detection means by an output signal of the fricative sound extraction means.