JPS5925237B2 - Speech segment determination method using speech analysis and synthesis method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for determining voiced sections, unvoiced sections, and silent sections of speech in a short time in a speech analysis and synthesis method.

In general, in speech analysis and synthesis systems, in order to compress speech information, voiced sections, unvoiced sections, and silent sections are determined, and information is extracted in the optimal and minimum amount for each section. It is taken from the method.

Therefore, how to determine this interval is an important issue in speech analysis and synthesis systems. In conventional speech analysis and synthesis equipment, for example, equipment using the PARCOR method, a residual signal is created by removing frequency spectrum envelope components such as formants from the audio signal, and a modified correlation function that is an autocorrelation function of the residual signal is created. After finding the maximum value and the first PARCOR coefficient kl, voiced or unvoiced is determined.

In practice, these processes are often performed using an electronic computer, but it takes a considerable amount of calculation time to obtain the residual signal and the modified correlation function. As part of efforts to speed up calculation processing, a method has been proposed to obtain a modified correlation function using a weighted moving average operation in which a digital filter is applied to the autocorrelation function of the audio waveform, but this method also takes a considerable amount of calculation processing time. At present, there is not much difference. The present invention efficiently, quickly, and accurately determines voiced, unvoiced, and silent sections of speech using a combination of an autocorrelation function of a speech waveform that has undergone simple preprocessing and a zero-crossing rate. This is to improve the calculation time problem.

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

FIG. 1 is a flowchart showing the outline of the process. In the same figure, 101 is audio waveform data, this waveform is divided into frames at a certain time interval (for example, 30 ms), and voiced, unvoiced, or silent is determined for each frame. Let's do it. In the process 102, the audio is clipped at the silent level, and a frame in which all sections within the frame are below the silent level is determined to be a silent section, and subsequent processing is not performed.

Ideally, this silent level is zero, but in reality it is set to a certain level (for example, ±2048
Set ±3) as the level integer type data. 103
The process is a process of counting the number of zero crossings.

The process will be explained with reference to FIG. Here, FIG. 2 shows an example of audio data, and in this example, the number of zero crossings is 7 for 22 pieces of data, and the zero crossing rate is 7/22. In unvoiced sections, this zero-crossing rate increases and holds one of the keys to determining voicedness and unvoicedness. When calculating this rate, there may be cases where the frame length is long and the frame break spans unvoiced and voiced periods. An example of this is shown in FIG. 3. In the case of this frame of data, it is thought that the left side of the frame spans an unvoiced section and the right side spans a voiced section. Speaking only in Japanese, about 70% are voiced sections, and the remaining 30% are voiced sections.
% is silent or silent section. It is difficult to determine whether such a frame is a voiced or unvoiced section. In the present invention, in such a case, it is determined that the segment is a voiceless section, the voiceless sound is emphasized, and the intelligibility of the voiceless consonant is prevented from deteriorating. For this purpose, when calculating the zero crossing rate, divide the frame into two,
The greater of the zero crossing rate in the first half of the frame and the zero crossing rate in the second half of the frame is adopted as the representative value within that frame.
104 is a process of determining whether the zero crossing rate is less than or equal to a threshold value for determining a silent section.

This is a value that can be determined as a silent section (for example, 1/
100) is a process of determining a frame having a value equal to or higher than a zero-crossing rate (for example, 1/3) that can be definitely determined to be an unvoiced interval to be an unvoiced interval.

Figure 4 shows the zero crossing rate Z (0) on the X axis, and the ratio W ('V ) - φ(T)/φ(0) 〇 Figure 4 shows each frame as Z(0.!-?
This is a graph for explaining classification into voiced (VOiced) sections or non-voiced (4) NVOiced) sections according to the relationship of F('Tt), and the process of 105 is similar to 401 in FIG.
corresponds to the area of (In Figure 4, the unvoiced section is (UV)
, a voiced section is (displayed as).If it is a voiced section, the autocorrelation function is calculated to find the fundamental frequency of the voice in that section, but the calculation process is long. In consideration of this, in order to reduce the number of calculations as much as possible, unvoiced sections are determined using a zero crossing rate as preprocessing. That is, if it is a completely silent or silent section, the autocorrelation function is not calculated because there is no need to find the fundamental audio frequency of that section. 106 is 105
This process is performed only on sections that have not been previously determined to be silent sections or voiceless sections, and reduces the influence of formants and eliminates the influence on the high-frequency component force width self-correlation function in sections where the audio signal is close to zero. This is a process for

A specific method will be explained with reference to FIG. FIG. 5 shows an audio signal within one frame. As shown in the figure, the frame is divided into three parts, and the value 503 is N% (for example, 30%) of the smaller value 502 of the maximum absolute value 501 of the first 1/3 section and the absolute value 502 of the second 1/3 section. Performs clipping processing. After this processing, the autocorrelation function φ(τ) is obtained in step 107, and when the value is extremely small (for example, when the maximum value of the autocorrelation function for 300 points of an integer type with ±2048 levels is 5 or less) ), the process of determining that section as a silent section is 108
It is. 109, from the maximum value φ('f!) of the autocorrelation function within the pitch frequency search section of the voice, r(7)
- Find φ('T!)/φ(O.

In general, F(T) is a somewhat large value if there is periodicity (
For example, it is known to take 0.4 or more). In the relationship between zero crossing rate Z(O and F(1) - φ('f!)/φ(0), the linear inequality F(
T) <AZ(0 (a is a statistically determined constant, for example, the part of 1.5 is determined to be an unvoiced section in step 110. 111 is shown in FIG. 4) Determine the unvoiced area of 403 In the process, an area 403 in which ?F(T)/φ(0) is less than a certain value (for example, 0.3) is determined to be an unvoiced section, and the fourth
This is a process of determining the remaining area 404 in the figure as a voiced section.

Figure 6 is an example of a graph showing the relationship between the autocorrelation function φ(τ) and the delay time τ. ) is the fundamental period of speech. As is clear from the above description, according to the present invention, voiced, unvoiced, and silent sections of the driving sound source component can be detected with high precision by determining the autocorrelation function and zero crossing rate of the audio waveform and combining them. can be achieved in a short time, and high-quality synthesized sounds can be obtained using this one-drive sound intensity component.

Furthermore, as a result, it is extremely effective in that it can be easily incorporated into the driving sound source signal analysis section of an existing speech analysis and synthesis device.

[Brief Description of the Drawings] Fig. 1 is a flowchart of a speech segment determination method using a speech analysis and synthesis method according to an embodiment of the present invention, Fig. 2 is a waveform diagram showing zero crossing points in a speech waveform, and Fig. 3 is a Figure 4 is a waveform diagram in which unvoiced and voiced sections are included in one frame, and the unvoiced and voiced sections are determined in sequence based on the flowchart in Figure 1. FIG. 5 is an explanatory diagram of preprocessing for removing high frequency components and the π influence of formants from an audio signal, and FIG. 6 is an explanatory diagram of a method for determining the pitch period.

Claims (1)

[Claims] 1. After applying a first level of clipping that removes hum components, etc. in the audio signal, the zero-crossing rate of the signal is determined, and from the value of the zero-crossing rate, certain silent sections and silent sections are determined in advance. is determined, and the audio signal is subjected to a second clipping process that reduces the influence of its high frequency components and formants for other sections. Then, the autocorrelation function φ(τ) is determined, and its maximum value φ(T ) and the value φ(O) of the autocorrelation function at zero delay time of the speech waveform Ψ(T), and the zero crossing rate are combined and cut at a certain threshold to determine the voiced section. A speech interval determination method using a speech analysis and synthesis method, characterized in that: 2 When voiced and unvoiced sections are mixed in a certain frame,
Claims characterized in that the frame is divided, and the largest value of the zero-crossing rate among the divided frames is replaced with the representative value of the zero-crossing rate of that frame, so as not to reduce the silent section. 2. A speech segment determination method using the speech analysis and synthesis method described in item 1.