JPH0832526A - Voice detector - Google Patents

Abstract

PURPOSE:To reduce an unpleasant sense of a reproduced voice by preventing missing of a voice frame due to noise/voice detection error and executing voice detection processing without giving an unpleasant sense to a recipient. CONSTITUTION:A VAD threshold level 1, that is, a voice detection threshold level 1 is equal to a prediction coefficient threshold level used to decide a noise region by a noise region decision device, and a VAD threshold level 2 and a VAD threshold level 3 are used to extract a region of higher frequency of occurrence than that of the VAD threshold level 1. Thus, they are set to decide a narrower region to be a noise (silence section). A voice/silence deciding device 35 sets a voice detection flag (u) to silence L when a mean value of prediction coefficients a1, a2 enters a specific range set by any of the VAD threshold levels 1, 2, 3 selected by a threshold level changeover device 41 among plural predetermined ranges and sets the flag to a voiced sound H in other cases. A final voice detection output (v) is obtained from the obtained result by using a hang-over processing circuit 36.

Description

Detailed Description of the Invention

The present invention relates to a voice detector used in a voice coding system.

[0002]

2. Description of the Related Art In a portable radio device or the like, in order to reduce power consumption during transmission, VOX (Voice Operate Swing) which transmits only when there is voice and suspends transmission when there is no voice
itch exchange) control is used, which can reduce the average power during transmission. In order to execute such a VOX function, a voice detector for detecting the presence / absence of a voice signal is required in the preceding stage of the transmission output circuit. One of the inventors of the present application previously proposed such a voice detector [see Japanese Patent Application Laid-Open No. 4-301930].

[0003]

However, in the prior art, as the background noise becomes larger because the prediction coefficient threshold for determining the voice / silence is fixed, the occurrence frequency distribution region of the prediction coefficient of the voice predicts the noise. Since the region is close to the coefficient occurrence frequency distribution region, a determination error that a voice segment is determined to be a noise segment occurs, and a voice segment is missing in the transmission output. The lack of the voice section not only makes the speaker uncomfortable, but also reduces the intelligibility of the conversation. Therefore, it cannot be said to be sufficient for voice detection in a noisy environment.

An object of the present invention is to provide a voice detector which eliminates the loss of voice signal due to VOX control in a noisy environment, which is a problem of the prior art, and reduces the discomfort of the reproduced voice on the receiving side. is there.

[0005]

To achieve this object, a speech detector according to the present invention comprises a first framer for framing a prediction coefficient obtained from an adaptive predictor of a speech coding apparatus, and An average value calculator for calculating the average value of the prediction coefficient in the frame section, and a prediction coefficient threshold value previously obtained from the occurrence distribution of the prediction coefficient and the average value are compared to generate a high occurrence frequency of the noise region occurrence frequency distribution. A noise region determiner for determining a noise region by finding a section located in the portion, and a uniform P
A second framing device for framing the CM signal; a power calculator for calculating a power average value of the uniform PCM signal only when the framed uniform PCM signal is determined to be in the noise region; Only when it is determined that the power average value of the uniform PCM signal outputs a determination result as to which region among a plurality of regions set by a certain number of determined power thresholds. A threshold switch for switching the voice detection threshold to a value higher than the prediction coefficient threshold when it is determined from the determination result that the power average value is in a high region, and the uniform PCM By determining whether the average value of the prediction coefficient is within a specific range set by the voice detection threshold value switched by the threshold value switcher to determine whether the signal is voiced or not. A voice / sound determiner that outputs a voice / silence flag indicating whether the segment is a voice segment or a silence segment, and outputs a voice detection output signal by performing an appropriate hangover process on the flag. And a hangover processing device that operates.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS As an embodiment, the present invention is a PHS (Personal Handyphone System) standard encoding system.
An example applied to 2 kb / s (kilobits / second) ADPCM is shown below. FIG. 2 is a block diagram of an ADPCM voice encoding device having a voice detection function to which the present invention is applied.
FIG. 1 is a block diagram showing an embodiment of a voice detector of the present invention.

First, the ADPCM coding apparatus shown in FIG. 2 will be described. Reference numeral 21 denotes a uniform PCM converter for converting a 64 kb / s μ-law PCM input signal into a uniform 13-bit PCM uniform PCM signal o. 22 is a uniform PCM converter 21
Is a subtractor that subtracts the prediction signal j, which is the output of the adaptive predictor 23, from the output o of the above to obtain the difference signal k. An adaptive quantizer 24 quantizes the difference signal k. 26 is 32k
It is an adaptive dequantizer that dequantizes b / s voice data and outputs a quantized difference signal m. 25 is the quantized difference signal m
And a predicted signal j and add the predicted signal j to output a reproduced signal n. Reference numeral 23 is an adaptive predictor that generates a prediction signal j from the reproduction signal n and the quantized difference signal m. 27 is a voice detector according to the present invention.

The 64 kb / s μ-law PCM input signal is converted into a uniform 13-bit PCM uniform PCM signal o.
Is converted by. The subtractor 22 subtracts the prediction signal j output from the adaptive predictor 23 from the uniform PCM signal o to obtain the difference signal k. This differential signal k is quantized by the adaptive quantizer 24, and 32 kb / s voice data is sent to the transmission line as the output of the ADPCM voice coding device. On the other hand, the adaptive dequantizer 26 adaptively dequantizes the voice data of 32 kb / s and outputs the quantized difference signal m. In the adder 25, the quantized difference signal m and the prediction signal j are added,
The reproduction signal n is output. The adaptive predictor 23 calculates the prediction coefficients a1 and a2 and uses them to generate the prediction signal j from the quantized difference signal m and the reproduction signal n. Prediction coefficient a calculated by the adaptive predictor 23 to generate the prediction signal j
1, a2 are coefficients for predicting a sample value at a certain time point with two adjacent sample values in the past, and the values thereof are different between a case of a speech signal with a large autocorrelation and a case of background noise with a small autocorrelation The distribution of occurrence is different in the area. The prediction coefficients a1 and a2 and the uniform PCM signal o are input to the voice detector 27 of the present invention. The voice detector 27 uses these signals to determine whether there is sound or no sound and outputs a voice detection output signal.

Next, the voice detector of the present invention will be described. FIG. 1 is a block diagram showing a configuration example of a voice detector of the present invention. Reference numerals 31 and 32 are first framing devices for framing the prediction coefficients a1 and a2 obtained from the adaptive predictor 23 into 5 msec, respectively. 33 and 34 are average value calculators that calculate the average value of each output of the first framing devices 31 and 32, respectively. 35 is an average value calculator 33, 34
Is a voiced / non-voiced discriminator which compares the output of the above with a given threshold value to make a voiced / non-voiced determination. Reference numeral 36 denotes a hangover processing device that performs a hangover process on the result of the sound / silence determiner 35. 37 is a second framing device for framing the uniform PCM signal o. Reference numeral 38 compares whether or not each output of the average value calculators 33 and 34 is included in the area A of the predetermined prediction coefficient threshold value as a noise area from the occurrence distribution of the prediction coefficients a1 and a2. , A noise area determiner that outputs a noise area signal t when it is determined to be included in the noise area A. Reference numeral 39 is a power calculator that calculates the average value of the power output from the second framing device 37. Reference numeral 40 denotes a voice detection threshold value given to the voice / non-voice determination device 35, that is, VAD (Voice Activity Detecti).
on) VAD threshold 1, VAD threshold 2, V
In order to switch the AD threshold value 3 according to the power level of the background noise, the output of the power calculator 39 is set to a preset power threshold value (Back Gr
bgnth_1, b used as sound noise threshold
gnth_2) to determine the magnitude of power. Reference numeral 41 designates a voice detection threshold value (VAD threshold value 1, VAD threshold value 2, V
This is a threshold value switching device for switching the AD threshold value 3).

First, the prediction coefficients a1 and a2 are framed by the first framers 31 and 32, and the average value calculator 3
At 3,34 the average value is calculated. From the noise area determiner 38, the power calculator 39 is used to determine whether the frame is in the noise area using the frame average values of the prediction coefficients a1 and a2.
And output to the power determiner 40. The power calculator 39 calculates the power average value (moving average value) of the equation (1) only when receiving the noise area signal t from the noise area determiner 38.

[Equation 1] bgnp = pbgnp × 0.95 + nbgnp × 0.05 (1) where, bgnp: average noise power value pbgnp: average power value of previous noise area frame nbgnp: average power value of current noise area frame

The power determiner 40 compares the output (bgnp) of the power calculator 39 with the power threshold for background noise power determination only when it receives the noise region signal t, and makes a determination according to, for example, equation (2). The threshold switch 41 is controlled to select the VAD threshold 1, VAD threshold 2 or VAD threshold 3 corresponding to the determination result as shown in FIG. 35 as a VAD threshold.

[Equation 2] Here, bgnth_n (n = 1, 2) is the power fixed as the threshold to the power determiner 40 in order to switch the VAD thresholds 1, 2, and 3 by the relationship of the following expression (3).

[Equation 3] bgnth_1 <bgnth_2 (3)

The VAD threshold 1, that is, the voice detection threshold 1, is equal to the prediction coefficient threshold for determining the noise region by the noise region determiner 38, and the VAD thresholds 2, V
The AD threshold value 3 is set so that a narrower area is determined as noise (silent section) in order to extract an area having a higher occurrence frequency than the VAD threshold value 1. The voiced / non-voiced determination unit 35 uses the threshold value switching unit 41 among a plurality of predetermined ranges in which the average values of the prediction coefficients a1 and a2 are as shown in FIG.
If it falls within a specific range set by the VAD threshold value 1, 2 or 3 selected in step 3, the voice detection flag u is set to silence (L).
Otherwise, set to voice (H) otherwise. The hangover processing device 36 is applied to the results obtained above.
Therefore, here, as an example, a hangover process of 100 msec is performed to obtain a final voice detection output signal v.

[0013]

As described above in detail, according to the present invention, the voice detection processing is executed without causing the listener to feel uncomfortable by preventing the voice frame from being dropped due to noise / voice detection error in a noisy environment. Therefore, the effect is extremely large.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a voice detector of the present invention.

FIG. 2 is a block diagram of an ADPCM speech coding apparatus having a speech detection function.

FIG. 3 is a characteristic diagram for explaining the operation of the present invention.

FIG. 4 is a characteristic diagram for explaining the operation of the present invention.

FIG. 5 is a characteristic diagram for explaining the operation of the present invention.

[Explanation of Codes] 31, 32 First Framer (for Prediction Coefficient) 33, 34 Average Value Calculator (for Prediction Coefficient) 35 Sound / Silence Determiner 36 Hangover Processing Device 37 Second Framer (For uniform PCM signal) 38 Noise region determiner 39 Power calculator 40 Power determiner 41 Threshold switching device t Noise region signal u Speech detection flag v Speech detection output signal 21 Uniform PCM converter 22 Subtractor 23 Adaptive predictor 24 adaptive quantizer 25 adder 26 adaptive dequantizer 27 speech detector j prediction signal k difference signal m quantized difference signal n reproduced signal o uniform PCM signal

Claims (1)

[Claims] 1. A first frame unit for framing a prediction coefficient obtained from an adaptive predictor of a speech coding apparatus, an average value calculator for calculating an average value of a frame section of the prediction coefficient, and the prediction coefficient. A noise region determiner that compares a prediction coefficient threshold previously obtained from the occurrence distribution of the above with the average value to find a section located in a high occurrence frequency portion of the noise region occurrence frequency distribution to determine the noise region, and a uniform A second framing device for framing the PCM signal; a power calculator for calculating a power average value of the uniform PCM signal only when the framed uniform PCM signal is determined to be in the noise region; Only when it is determined that the average power value of the uniform PCM signal is in a plurality of regions set by a certain number of predetermined power thresholds A determining device, and a threshold value switching device for switching the voice detection threshold value to be higher than the prediction coefficient threshold value when it is determined by the determination result that the power average value is in a high region, By determining whether the average value of the prediction coefficient is within a specific range set by the voice detection threshold value switched by the threshold value switcher to determine whether the uniform PCM signal is voiced or not. A voice / non-voice deciding device for outputting a voice / non-voice flag indicating whether the period is a voice period or a silence period, and a voice detection output signal by performing an appropriate hangover process on the flag. A voice detector having a hangover processing device.