JPH04115299A - Method and device for voiced/voiceless sound decision making - Google Patents

Abstract

PURPOSE:To decide whether a voice signal is voiced or voiceless by predicting the state of the voice signal and adapting a threshold value for deciding whether or not there is a pitch and a threshold value for deciding the stationary/nonstationary state of a spectrum to the state of a voice which is inputted next. CONSTITUTION:A pitch presence/absence decision means 11 decides whether or not the input voice signal has a pitch by using the threshold value. Then a spectrum stationary/nonstationary state decision means 12 decides whether or not the spectrum of the input signal has variation from the spectrum in the last frame, namely, the stationary or nonstationary state by using the threshold value. Then a voiced/voiceless decision means 13 decides that the input signal which has no pitch and a stationary spectrum is a voiceless signal and other signals are voiced signals. Then a voice section predicting means 14 decides the state of the current voice signal from those decision results and predict the state of the next frame. Then a threshold value setting means 15 sets the threshold value suitable for the pitch presence/absence decision making and the threshold value suitable for the stationary/nonstationary spectrum decision making according to the prediction result.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a method and apparatus for determining voice presence/absence for use in digital communications, voice mail, packet communications, and the like.

BACKGROUND OF THE INVENTION FIG. 3 shows an example of a conventional method for determining voice presence/absence. In FIG. 3, first, an audio signal is input at regular time intervals (hereinafter referred to as frames) (step 1), and it is determined whether or not the input audio signal has a pitch (step 2). If there is no pitch, it is further determined whether the spectrum of the input audio signal is stationary or non-stationary over a plurality of previous frames (step 3). If there is no pitch and the spectrum is stationary, the input audio signal is determined to be silent (step 5), and anything else is determined to be sound (step 4).

In this way, even with the above-mentioned conventional speech presence/non-speech determination method, a noise signal with no pitch and a stationary spectrum is determined to be silent,
Even if a consonant whose spectrum changes rapidly or a vowel whose spectrum is stationary, it can be determined that it is voiced.

Problems to be Solved by the Invention However, in the above-mentioned conventional voice presence/absence determination method,
If there is no pitch and the spectrum is stationary, it is determined to be silent, so even if the spectrum is non-stationary, voiceless consonants with no pitch or voiced consonants whose pitch is difficult to determine are mistakenly determined to be silent. There was a problem.

The present invention solves such conventional problems, and provides a highly accurate speech presence/non-speech determination method and device that can correctly determine the above-mentioned voiceless consonants and voiced consonants as speech. The purpose is to

Means for Solving the Problems In order to achieve the above-mentioned object, the present invention provides a determination result of whether or not there is a pitch in an input audio signal, a determination result of whether the spectrum is stationary or non-stationary, and a determination result of whether there is voice or not. Based on the judgment result, predict the state of the next input audio signal, such as whether it is a silent section, a consonant section, or a vowel section, and based on this prediction, whether there is a pitch or not. The threshold value for determining whether the spectrum is stationary or unsteady is adapted to the state of the next input voice to determine whether the voice is voiced or not. It is something.

Effects The present invention has the following effects due to the above structure. In other words, the state of the audio signal of the next frame is predicted based on the pitch non-judgment result of the current frame, the steady/unsteady spectrum determination result, and the presence/absence determination result, and the pitch presence/absence determination is performed according to the prediction result. By appropriately setting the threshold value for determining whether the spectrum is stationary or non-stationary, it is possible to determine whether the voice is present or not according to the state of the voice. This has the effect of increasing the accuracy of.

Embodiment FIG. 1 shows the configuration of an embodiment of the present invention. In FIG. 1, reference numeral 11 denotes a pitch presence/absence determining means, which determines whether or not the input audio signal has a pitch. 12
is a spectrum stationary/unstationary determining means, which determines whether the spectrum of the input audio signal is stationary or unsteady. Reference numeral 13 denotes a voice presence/absence determination means, which determines whether or not there is voice in the input audio signal based on the results determined by the pitch presence/absence determination means 11 and the spectral stationary non-stationary determination means 12 . 14 is a speech interval prediction means;
Based on the respective determination results of the pitch presence/absence determination means 11, the spectral stationary non-stationary determination means 12, and the voice presence/absence determination means 13, the current state of the sound is determined, and the state of the sound of the next frame is predicted. 15 is a threshold value setting means;
Based on the results obtained by the speech interval prediction means 14, a threshold value adapted to the determination by the pitch presence/absence determination means 11 and the spectral stationary non-stationary determination means 12 is set.

Next, the operation of the above embodiment will be explained using FIG. 2. In FIG. 2, first, an A/D converted audio signal is input frame by frame (step 21), and the pitch presence/absence determining means 11 determines whether or not the input audio signal has a pitch (step 22). Various methods can be considered for this determination, but in this case, if the parameter for determining the presence or absence of pitch is less than or equal to a threshold value, it is determined that there is no pitch, and if it is greater than or equal to the threshold value, it is determined that there is a pitch. Next, the spectrum stationary/unsteady determination means 12 uses a threshold to determine whether there is a variation in the spectrum of the input signal with the spectrum of the previous frame, that is, whether it is stationary or unsteady ( Step 23). In this case, if the variation is less than or equal to the threshold value, it is determined to be steady, and if it is greater than or equal to the threshold value, it is determined to be unsteady. Then, the voice presence/absence determining means 13 determines as silence the input signal for which there is no pitch and the spectrum is determined to be stationary (step 25).If there is a pitch or the spectrum is unsteady, it is determined to be voice. do(
Step 24).

Next, the audio section prediction means 14 determines the current state of the audio signal from these determination results and predicts the signal state of the next frame (step 26). That is,
If it is determined to be silent, it is determined to be a silent section, and if there is a sudden change in the spectrum after the silent section, it is determined that it is a rising section of speech, that is, a consonant section, and the next frame after the rising section is a frame with a pitch. If it continues, it is determined that it is a vowel section of the voice, and if there is a change in the spectrum after the vowel zone continues, and frames with pitch immediately follow in order, it is determined that it is a transition section of voice, and if there is no pitch, it is determined that there is no sound. It is determined that it is an interval.

Then, in the threshold value setting means 15, a threshold value adapted to determine the presence or absence of a pitch and a threshold value adapted to determine whether the spectrum is stationary or unsteady are set, respectively, based on the prediction result of step 26 (step 27). . In other words, when the input signal is determined to be a silent interval, the voice interval prediction means 14 predicts that the next input signal is a silent signal or a voice rising signal, and determines whether the spectrum is stationary or non-stationary. The threshold value is lowered to increase the sensitivity to the spectrum, making it easier to determine the rising edge of the voice. Additionally, if the input signal is determined to be a consonant interval, the next input signal is predicted to be a vowel interval, and the pitch threshold is lowered to increase the pitch sensitivity and the spectral threshold is raised. Lowers sensitivity to changes in the spectrum, making it easier to identify vowels. In addition, if the input signal is determined to be a vowel interval, the next input signal is predicted to be a voice transition interval or a silent interval, and the spectral threshold is lowered to increase the sensitivity to spectral changes. In addition to making it easier to distinguish between crossing and silence,
Since a vowel follows after a large change in the spectrum, the sensitivity to pitch is increased to make it easier to determine the vowel.

In this way, according to the above embodiment, the following is calculated based on the determination result of whether or not there is a pitch in the input audio signal, the determination result of whether the spectrum is stationary or unsteady, and the determination result of whether voice is present or absent. It predicts whether the signal is a silent section of speech, a consonant section at the beginning of speech, a vowel section of speech, or a transition section of speech, and based on this prediction, the pitch is determined. The threshold value for determining whether the spectrum is stationary or non-stationary, and the threshold value for determining whether the spectrum is stationary or non-stationary, are changed as appropriate depending on the state of the next input voice, so it is easy to determine whether the voice is voiced or not. can be performed accurately, and the accuracy of voice presence/absence determination can be improved.

Effects of the Invention As is clear from the above-mentioned embodiments, the present invention provides a determination result as to whether or not there is a pitch in an input audio signal, a determination result as to whether the spectrum is stationary or non-stationary, and a determination result as to whether there is voice or not. To predict the state of the audio signal such as whether the next signal is a silent section, a consonant section, or a vowel section from and the threshold for determining whether the spectrum is stationary or non-stationary, respectively, are changed to adapt to the state of the next input voice to determine whether the voice is voiced or not. The present invention has the advantage that it is possible to accurately determine whether a voice is uttered or not, and it is possible to realize a highly accurate method and device for determining whether a voice is uttered or not.

[Brief explanation of drawings]

FIG. 1 is a schematic block diagram of a voice presence/absence determination device according to an embodiment of the present invention, FIG. 2 is a flowchart for explaining the operation of the device, and FIG. It is a flow chart for explanation. DESCRIPTION OF SYMBOLS 11... Pitch presence/absence determination means, 12... Spectrum steady/unsteady determination means, 13... Voice presence/absence determination means, 14... Voice segment prediction means, 15... Threshold value setting means. Name of agent: Patent attorney Akira Okaji and two others

Claims (2)

[Claims] (1) Determine whether the input audio signal has a pitch or not, determine whether the spectrum is stationary or unsteady, determine whether there is audio or not based on these results, and The state of the next input audio signal is predicted from the determination result, and based on this prediction, a threshold value for determining the presence or absence of pitch and a threshold value for determining whether the spectrum is stationary or non-stationary are respectively input. A voice presence/absence determination method is configured to be set in accordance with the state of the voice being played. (2) Pitch presence/absence determination means for determining whether the input audio signal has a pitch or not; spectrum stationary/unsteady determination means for determining whether the spectrum is stationary or unsteady; A voice presence/no-sound determination means for determining whether there is voice or not; a voice interval prediction means for predicting the state of the next input voice signal from the respective determination results of the respective means; and a pitch prediction means based on the prediction. A voice presence/absence determination device comprising a threshold value setting means for setting a threshold value adapted to determine the presence or absence of a spectrum and a threshold value adapted to determine whether a spectrum is stationary or unsteady.