Speech detection apparatus and speech detection method

Abstract

To provide a device and a method for voice detection which can detect voice with a small memory, an inexpensive CPU, and a simple algorithm and can be realized by an inexpensive device.SOLUTION: A voice detection device 10 includes a band division part for dividing a sound signal from an environment into predetermined frequency bands, a power determination part for determining whether a power value of the lowest band is maximum and exceeds a predetermined threshold, and a voice determination part for determining that the sound signal is a voice when the fact that the power value exceeds the predetermined threshold appears at a predetermined frequency within a predetermined time.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a voice detection device and a voice detection method. [Background technology]

[0002] Some voice communication devices compress voice data for communication. When communication speeds are slow, real-time communication is achieved by increasing the compression rate. Wireless devices tend to have slow communication speeds and can only transmit a small amount of data in real time. To reduce the amount of data, some devices detect voice activity and reduce the amount of data sent when there is no voice activity, making effective use of the limited communication speed.

[0003] If voice detection is performed based only on the power value of the input to the communication device, noise will also be judged to be voice, so a voice detection device should be used to distinguish between voice and noise. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2018-180482 [Patent Document 2] Japanese Patent Publication No. 2022-32721 Summary of the Invention [Problem to be solved by the invention]

[0005] The voice detection device described in the above-mentioned Patent Document 1 converts an input acoustic signal into the frequency domain using FFT (Fast Fourier Transform), calculates a power spectrum, extracts the fundamental frequency from this spectrum, and uses it to determine voice detection. The voice detection device described in Patent Document 1 requires information processing to perform complex voice processing, similar to the technology used to process voice signals, and requires large-capacity memory, a CPU with high computing capabilities, etc., making the device expensive.

[0006] The voice detection device described in the above-mentioned Patent Document 2 uses a histogram of the signal level of an input sound signal to determine voice detection based on fluctuations in level. The voice detection device described in Patent Document 2 may possibly determine that noise is voice if the fluctuation in level is small.

[0007] The present invention solves the problems of the conventional technology described above, and aims to provide a voice detection device and method that can determine voice using inexpensive resources such as a small-capacity memory and an inexpensive CPU, and a simple algorithm. [Means for solving the problem]

[0008] A first aspect of the present invention is a voice detection device that determines voice contained in an input sound signal, characterized by having a band division unit that divides the input sound signal into a plurality of frequency bands, a power determination unit that determines whether the power value of the lowest frequency band among the divided bands is the largest and exceeds a predetermined threshold, and a voice determination unit that determines whether the signal is voice based on the output of the power determination unit.

[0009] It is preferable that the band dividing section divides the signal by combining two divisions using a low-pass filter and a high-pass filter.

[0010] Another aspect of the present invention is a voice detection method for determining voice contained in an input sound signal, characterized by comprising a band division step for dividing the input sound signal into a plurality of frequency bands, a power determination step for determining whether the power value of the lowest frequency band among the divided frequencies is the largest and exceeds a predetermined threshold, and a voice determination step for determining whether the signal is voice based on the output of a power determination unit. [Effects of the Invention]

[0011] Taking advantage of the fact that speech has high power in the low frequency bands and that noise has power distributed across all bands, speech is determined by whether the power in the low frequency bands is greater than the power in other bands.This means that speech detection is possible with little memory, an inexpensive CPU, and a simple algorithm, and can be implemented on inexpensive equipment. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a diagram showing the configuration of a voice detection device according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram illustrating a band dividing unit according to the embodiment. [Figure 3] FIG. 3 is a diagram illustrating the band 2 division unit. [Figure 4] FIG. 4 is a diagram illustrating a power determination unit according to the embodiment. [Figure 5] FIG. 5 is a diagram illustrating the relationship between the frequency and power of speech and noise. [Figure 6] FIG. 6 is a diagram illustrating the determination operation of the voice determination unit according to the embodiment. [Figure 7] FIG. 7 is a flowchart illustrating the operation of the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0013] An embodiment of the present invention will now be described with reference to the drawings. Fig. 1 is a diagram showing the configuration of a voice detection device 10 according to an embodiment of the present invention, which includes an A / D conversion unit 11 that converts an input acoustic signal from the environment into a digital signal, a band division unit 12 that divides the frequency bands of the acoustic signal converted into a digital signal, a power determination unit 13 that determines whether the power value of the lowest band among the divided bands is the largest and exceeds a predetermined threshold, and a voice determination unit 14 that determines whether there is voice based on the output of the power determination unit.

[0014] 2 is a diagram illustrating the band dividing unit 12 according to the embodiment. The band dividing unit 12 has a function of dividing the acoustic signal, which has been converted into a digital signal by the A / D conversion unit 11, into a plurality of frequency bands by combining two divided bands, with the lowest frequency band being a predetermined band. The predetermined band is the band in which the audio power value is assumed to be concentrated. For example, if the audio power value is set to 0 to 1000 Hz and the sampling frequency of the A / D conversion unit 11 is 8000 Hz, combining three two-band divisions results in four frequency bands: Band 1 from 0 to 1000 Hz (however, the lowest frequency can be set to the audible band), Band 2 from 1000 to 2000 Hz, Band 3 from 2000 to 3000 Hz, and Band 4 from 3000 to 4000 Hz. The divided acoustic signals are output to the power determining unit 13 at the subsequent stage.

[0015] 3 is a diagram illustrating the band 2 division unit 121. The band 2 division unit 121 has a function of dividing the acoustic signal into two frequency bands by using a low-pass filter and a high-pass filter.

[0016] FIG. 4 is a diagram illustrating the power determination unit 13 of the embodiment, and includes a power comparison unit 131 and a maximum value selection unit 132. The power determination unit 13 has the function of outputting "1" when the power value of the lowest frequency band, in this case the frequency band from 0 Hz to 1000 Hz, from the power values ​​of the frequency bands divided by the band division unit 12 is higher than the power values ​​of the other frequency bands and is higher than a predetermined threshold A, for example, -40 dB when the maximum sound pressure is 0 dB in the case of a normal speaking voice, and outputting "0" when the power value of the lowest band is lower than the power values ​​of the other frequency bands or is lower than the predetermined threshold A. The power comparator 131 outputs "1" if the input power value is higher than the reference power value, and outputs "0" if it is lower. The maximum value selection unit 132 outputs the maximum power value from among the power values ​​of the frequency bands divided by the band division unit 12, excluding the lowest frequency band. The power comparison unit 131 can be realized by using a comparison function of a CPU that compares an input power value with a reference power value. The maximum value selection unit 132 can be realized by using a comparison function of a CPU that compares a plurality of input power values.

[0017] The principle of voice determination will be explained with reference to Figure 5. Figure 5(a) shows the frequency characteristics of voice, and (b) shows the frequency characteristics of noise, etc., with the horizontal axis representing frequency and the vertical axis representing power. While voice has its power concentrated below 1000 Hz, noise and other sounds have their power distributed across the entire frequency range, making them distinguishable from voice.

[0018] 6 is a diagram illustrating the voice determination unit 14 according to the embodiment. The unit is provided with 16 shift registers and a number comparison unit 141, and has a function of storing the output of the power determination unit 13 in the shift registers and outputting voice detection when the sum of the values ​​in the shift registers exceeds a predetermined threshold value B. The number comparison unit 141 outputs "1" if the input value is greater than the reference value, and outputs "0" if it is less than the reference value.

[0019] For example, the output of the power determination unit 13 is sampled every 10 milliseconds and stored in a shift register. Speech is determined based on a total of 16 "1's," that is, 10 milliseconds x 16 = 160 milliseconds.

[0020] 7 is a flowchart showing the above-mentioned voice detection operation. If the power value of the lowest frequency band is higher than the power values ​​of the other frequency bands every 10 milliseconds, a "1" is entered into the shift register; if not, a "0" is entered (step S1).

[0021] A determination is made based on the total number of "1"s in the shift registers over a period of 160 milliseconds (step S2). This determination is made based on the total number of "1"s in the 16 shift registers, and if the total number of "1"s is 8 or more in step S2, it is determined to be voice and voice detection is output (step S3).

[0022] (Explanation of effect) The above-described embodiment can be understood as follows. The voice detection device 10 judges the voice contained in an input sound signal, and includes a band division unit 12 that divides the input sound signal into a plurality of frequency bands, a power judgment unit 13 that judges whether the power value of the lowest frequency band among the divided bands is the largest and exceeds a predetermined threshold, and a voice judgment unit 14 that judges whether it is voice based on the output of the power judgment unit.

[0023] With this configuration, the determination of whether or not a signal is voice can be achieved by using a shift register and counting the data in that register, which requires a small amount of memory and calculations, making it possible to realize a voice detection device using inexpensive hardware.

[0024] Furthermore, the band division unit 12 can also be realized by a Fourier transform using an FFT, but since the number of bands into which the frequency is divided is small, it can be realized by a combination of a low-pass filter and a high-pass filter, which has the advantage of not requiring high calculation capabilities. Another advantage is that a band filter does not need to be used. If an FIR (Finite Impulse Response) filter is used for the process of dividing the band into two, the filter coefficients can be odd numbers, which reduces the amount of calculation processing. Another advantage is that the frequency separation characteristics are good.

[0025] Furthermore, in the embodiment of the present invention, since the determination is made based on whether the power value of the lowest band is the maximum and exceeds a predetermined threshold, voice detection is possible using a simple algorithm and few resources.

[0026] In the present invention, it is possible to determine that an input acoustic signal is not a voice signal in contrast to noise.

[0027] This invention can detect voice. Therefore, by transmitting only voice using a handheld wireless device, data communication volume and battery consumption can be reduced. In addition, by combining a voice detection device with a voice recording device, it is possible to save only voice, reducing memory requirements and extending recording time.

[0028] The present invention is suitable for use in voice devices that distinguish voice from noise and perform gain control in environments where there are not many sound sources and where meaningless noise is present, such as voice devices for communicating between the inside and outside of an elevator, or voice devices for communicating in a house or office, etc. Even if a low-frequency sound source, an alarm sound, etc. is determined to be voice, the effect of the present invention, which distinguishes voice from noise, is not diminished. [Explanation of symbols]

[0029] 10. Voice detection device 11 A / D conversion section 12 Band division section 121 Band 2 division section 13 Power judgment section 131 Power comparison section 132 Maximum value selection section 14. Audio detection unit 141 Quantity comparison unit

Claims

1. A voice detection device that determines voice contained in an input sound signal, a band dividing unit that divides an input acoustic signal into predetermined frequency bands; a power determination unit that determines whether the power value of the lowest band among the divided bands is the largest and exceeds a predetermined threshold; a voice determination unit that determines whether an output from the power determination unit is voice; and The voice determination unit a number determination unit that determines whether the output of the power determination unit appears at a predetermined frequency within a first time period; A voice detection device characterized by:

2. 2. The voice detection device according to claim 1, the band dividing unit divides the frequency band using a low-pass filter and a high-pass filter; A voice detection device characterized by:

3. 3. The voice detection device according to claim 1, the power determination unit includes a determination means for outputting a message indicating that "audio power has appeared" when the power value of the lowest band among the frequency bands divided by the band division unit is the largest and exceeds a predetermined threshold value; the voice determination unit stores the output of the power determination unit, and outputs voice detection when the output of the power determination unit exceeds a predetermined frequency within the first time period. A voice detection device characterized by:

4. A speech detection method for determining speech contained in an input acoustic signal, comprising: a band division step of dividing an input acoustic signal into predetermined frequency bands; a power determination step of determining whether the power value of the lowest band among the divided bands is the largest and exceeds a predetermined threshold; a voice determination step for determining whether or not the output of the power determination step is voice; and The voice determination step includes: a number determining step for determining whether an output from the power determining step appears at a predetermined frequency within a first time period; having A speech detection method comprising:

5. 5. The speech detection method of claim 4, further comprising: the band dividing step divides the frequency band using a low-pass filter and a high-pass filter; A speech detection method comprising:

6. 6. A speech detection method according to claim 4 or 5, comprising: the power determination step includes a determination means for outputting a signal indicating that "audio power has appeared" when the power value of the lowest band among the frequency bands divided in the band division step is the largest and exceeds a predetermined threshold value; the voice determination step stores an output of the power determination step, and outputs voice detection when the power determination step stores an output exceeding a predetermined frequency within the first time period; A speech detection method comprising:

Images