Voice detection method for helicopter voice processing system

Abstract

The invention belongs to the technical field of voice processing, and particularly relates to a voice detection method for a helicopter voice processing system. According to the method, the voice signal within a certain time is cached, and the steps of high-frequency filtering, framing, short-time Fourier transform, Mel filter processing, logarithm operation, discrete cosine transform and the like are carried out on the voice signal, so that the characteristics of the voice signal can be extracted more accurately. Compared with a traditional method based on a simple threshold, the method is finer, and the accuracy of noise recognition is improved. Due to the fact that noise signals within 500 milliseconds can be effectively removed, hearing experience of users is improved, hearing fatigue of the users is relieved, and comfort and working efficiency are improved. According to the voice detection method, the defects of an original digital squelch algorithm are overcome, the trailing noise caused by too long squelch turn-off delay in the voice is greatly weakened, the voice quality is improved, and it is guaranteed that the voice state is clear and no trailing noise exists in the whole machine flight state.

Description

Technical Field

[0001] The invention belongs to the technical field of speech processing, and in particular relates to a speech detection method for a helicopter speech processing system. Background Art

[0002] During the communication process of the helicopter, each channel will generate noise, such as the noise generated by the aircraft engine, the power supply interference generated by the converter, the pumping noise generated by various pressure pumps of the aircraft hydraulic system, and the noise generated by various equipment due to electromagnetic compatibility issues. In response to these noises, the digital voice processing system on the original helicopter added a noise reduction algorithm to each voice channel. This algorithm sets a threshold. When the threshold is exceeded, the input signal is considered to be voice and the signal is sent to the system. When the threshold is not reached, the input signal is considered to be noise and the signal is turned off. Since this method is relatively rough in identifying noise, in order to avoid the phenomenon of accidental voice shutdown (dropped words in the call), a delayed shutdown solution is usually used, and the delay time cannot be less than 500ms; however, the noise signal within 500ms cannot be removed, resulting in a period of noise being heard in the headset before the voice is turned off; therefore, a voice detection method is urgently needed to solve the problem that the noise signal within 500ms cannot be removed, resulting in noise being heard in the headset before the voice is turned off.

[0003] In view of this, the present invention is proposed. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art and provide a speech detection method for a helicopter speech processing system.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A speech detection method for a helicopter speech processing system, the specific steps are as follows:

[0007] Step 1: Cache the voice signal for a certain period of time and then send the signal data to the back end;

[0008] Step 2: filter the high-frequency part of the voice signal with a high-pass filter to emphasize the high-frequency component of the voice;

[0009] Step 3: Frame the voice signal obtained in step 2. Since the subsequent signal needs to be convolved in the frequency domain, a Hamming window is required to prevent sidelobe leakage. The voice data is divided into n frames. The framing process is y[n]=w[n]x[n], where x[n] is the voice data of each frame and w[n] is the window function of each frame. The window function is Wherein, L is usually an integer between 12 and 16;

[0010] Step 4: Perform short-time Fourier transform on each frame of the voice signal obtained in step 3. Where, j = -1, e is the base of natural logarithm, m is the frequency sampling point, m = 0, 1, 2..., N-1, Y[m] is the mth sampling output;

[0011] Step 5: The voice signal obtained in step 4 is filtered through a Mel filter to obtain energy functions of different frequencies; the values ​​of the Mel filter corresponding to voice signals of different frequencies are different, according to the formula And the following formula:

[0012]

[0013] , where mal represents the number of Mel filters, f() represents the interval frequency, and k represents the number of filters. Based on the obtained Mel matrix H, the data Y[m] after Fourier transformation is filtered by the Mel filter, and the data energy function E(m) = |Y(m)| 2 H;

[0014] Step 6: Perform logarithmic operation on the energy output by each filter.

[0015]

[0016] Step 7: Perform discrete cosine transform on the logarithmic signal to obtain the corresponding Mel cepstral coefficient, that is, (L is usually 12-16);

[0017] Step 8: Compare the obtained Mel cepstral coefficients with the noise coefficients in the vector table. The algorithm considers the first 5 frames of data collected at startup to be all noise and stores this noise feature in the vector table as the initial noise data, recorded as N1. The second 5 frames of data are subsequently cached and set to N2, which is considered to be the second noise data. According to T=N+k*(NN min ) method to determine the threshold value, where T is the threshold value, k is the coefficient, and different parameters are used to set different sensitivities, including 0, 1, 2, and 3, respectively. N is the current noise value, and N min is the minimum noise value; after obtaining the Mel coefficient, when the Mel cepstral coefficient is much larger than T, it is confirmed as speech, and when no significant mutation occurs, it is considered as noise, and this is used as the N value to update the threshold T; if it does not meet the requirements, it means that the input is speech, and if it meets the requirements, it means that the input is noise. At this time, the squelch gate control signal is output to close the squelch gate.

[0018] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0019] 1. Improve the accuracy of noise recognition: By caching the voice signal for a certain period of time and performing high-frequency filtering, framing, short-time Fourier transform, Mel filter processing, logarithmic operation and discrete cosine transform, the characteristics of the voice signal can be extracted more accurately. This method is more sophisticated than the traditional method based on simple thresholds and improves the accuracy of noise recognition.

[0020] 2. Enhanced real-time performance: This method can complete signal processing and noise identification in a relatively short time, avoiding the 500 millisecond delay problem in the traditional delay shutdown solution. This can remove noise in time before the voice signal appears, improving the quality of real-time communication.

[0021] 3. Improve user experience: Since noise signals within 500 milliseconds can be effectively removed, the noise heard by pilots and other passengers in the headphones is greatly reduced. This not only improves the user's auditory experience, but also reduces their auditory fatigue, improves comfort and work efficiency.

[0022] 4. Adaptability to various noise environments: This method can cope with various noise sources generated during helicopter communications, including engine noise, power supply interference, hydraulic system noise, and noise caused by electromagnetic compatibility issues. This wide adaptability enables the detection method to maintain good detection performance in different flight environments and conditions.

[0023] 5. Improve communication quality: Through more sophisticated noise recognition and processing, the impact of noise on communication is reduced, and the clarity and intelligibility of voice signals are improved. This is particularly important for mission-critical communications, especially in emergency situations, where clear communication can greatly improve the safety and efficiency of operations.

[0024] In summary, the speech detection method of the present invention makes up for the shortcomings of the original digital squelch algorithm, greatly reduces the tail noise in the voice caused by the long squelch shutdown delay, improves the voice quality, and ensures that the voice state of the entire aircraft is clear when in flight without the problem of tail noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings are incorporated in and constitute a part of this specification and, together with the description, serve to explain the principles of the present invention.

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0027] Figure 1 is a processing flow chart of the speech detection method of the present invention;

[0028] Figure 2 This is a schematic diagram of a collection chip in a circuit according to an embodiment of the present invention;

[0029] Figure 3 is a schematic diagram of a processor in a circuit according to an embodiment of the present invention;

[0030] Figure 4 Schematic diagram of a hardware printed circuit board according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are only examples consistent with some aspects of the present invention as detailed in the appended claims.

[0032] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0033] Example

[0034] See also Figure 1 As shown, this embodiment provides a voice detection method for a helicopter voice processing system, and the specific steps are as follows:

[0035] Step 1: The cycle of different digital voice systems is 1ms or 5ms respectively. This method needs to process 10ms of data at a time. Due to the different cycles of each system, a buffer needs to be added. When enough 10ms of data is stored, the signal data is sent to the back end.

[0036] Step 2: Filter the high-frequency part of the voice signal with a high-pass filter (3400 Hz) to emphasize the high-frequency component of the voice;

[0037] Step 3: Frame the voice signal obtained in step 2. Since the subsequent signal needs to be convolved in the frequency domain, a Hamming window is required to prevent sidelobe leakage. The system sampling rate is 16KHz. For 10ms of data, there are 160 sampling points. The voice data is divided into n frames. The framing process is y[n]=w[n]x[n], where x[n] is the voice data of each frame and w[n] is the window function of each frame. The window function is Wherein, L is usually an integer between 12 and 16;

[0038] Step 4: Perform short-time Fourier transform on each frame of the voice signal obtained in step 3. Where j = -1, e is the natural logarithm base, m is the frequency sampling point, m = 0, 1, 2..., N-1, Y[m] is the mth sampling output, and since there are 160 sampling points, m is 160;

[0039] Step 5: The voice signal obtained in step 4 is subjected to a Mel filter to obtain energy functions of different frequencies; different frequency values ​​corresponding to the Mel filter are different. The existing Mel filter is used, and its upper scale is set to 200 Hz, and the lower scale is set to 4000 Hz at most. 26 filters are set, and the formula is as follows: And the following formula:

[0040]

[0041] Among them, mal represents the number of Mel filters, f() represents the interval frequency, and k represents the number of filters. According to the obtained Mel matrix H, the data Y[m] after Fourier transformation is filtered by the Mel filter. The data energy function E(m)=|Y(m)| 2 H;

[0042] Step 6: Perform logarithmic operation on the energy output by each filter.

[0043]

[0044] Step 7: Perform discrete cosine transform on the logarithmic signal to obtain the corresponding Mel cepstral coefficient, that is, (L is usually 12-16);

[0045] Step 8: Compare the obtained Mel cepstral coefficients with the noise coefficients in the vector table. The algorithm considers the first 5 frames of data (50ms) collected at startup to be all noise and stores this noise feature in the vector table as the initial noise data, recorded as N1. The second 5 frames of data cached subsequently are set to N2, and this data is considered to be the second noise data. According to T=N+k*(NN min ) method to determine the threshold value, where T is the threshold value, k is the coefficient, and different parameters are used to set different sensitivities, including 0, 1, 2, and 3, respectively. N is the current noise value, and N min is the minimum noise value; after obtaining the Mel coefficient, when the Mel cepstral coefficient is much larger than T, it is confirmed as speech, and when no significant mutation occurs, it is considered as noise, and this is used as the N value to update the threshold T; when noise is identified, the squelch gate switch is controlled to close the squelch gate.

[0046] Considering that this voice detection method needs to be transplanted into various types of hardware circuits, a typical circuit is designed, with CS320AIC3106 as the acquisition chip and CS32F407ZG as the processor. This circuit requires the input signal size to be 100-120mV. The hardware circuit is shown in the figure. The I2S interface is used for communication between chips; see Figure 2-3 As shown, 3D1 is the domestic ARM chip CS32F407ZG, 3D2 is the domestic voice acquisition chip CS320AIC3106, 3C123, 3C124, 3C66, 3C67 are coupling capacitors, 3C1, 3C2, 3C3, 3C4, 3C5, 3C6, 3C7 are bypass filter capacitors, see Figure 4 shown.

[0047] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention.

[0048] It should be understood that the present invention is not limited to what has been described above and that various modifications and changes may be made without departing from its scope. The scope of the present invention is limited only by the appended claims.

Claims

1. A speech detection method for a helicopter speech processing system, characterized in that: The specific steps are as follows: Step 1: Cache the voice signal for a certain period of time and then send the signal data to the back end; Step 2: Filter the high frequency part of the voice signal to emphasize the high frequency component of the voice; Step 3: Frame the voice signal obtained in step 2. Since the subsequent signal needs to be convolved in the frequency domain, a Hamming window is required to prevent sidelobe leakage. Step 4: Perform short-time Fourier transform on each frame of the voice signal obtained in step 3; Step 5: The voice signal obtained in step 4 is filtered through a Mel filter to obtain energy functions of different frequencies; Step 6: Perform logarithmic operation on the energy output by each filter; Step 7: Perform discrete cosine transform on the logarithmic signal to obtain the corresponding Mel-frequency cepstrum coefficients; Step 8: Compare the obtained Mel-frequency cepstrum coefficients with the noise coefficients in the vector table. If they do not match, it means that the input is speech. If they match, it means that the input is noise. At this time, a squelch gate control signal is output to close the squelch gate.

2. The speech detection method according to claim 1, characterized in that: In step 2, filtering is performed using a high pass filter.

3. The speech detection method according to claim 1, characterized in that: In step 3, the speech data is divided into n frames, and the framing process is y[n]=w[n]x[n], where x[n] is the speech data of each frame, and w[n] is the window function of each frame.

4. The speech detection method according to claim 3, characterized in that: The window function is 0≤n≤L-1, where L is usually an integer between 12 and 16.

5. The speech detection method according to claim 3, characterized in that: In step 4, the short-time Fourier transform is Where j = -1, e is the natural logarithm base, m is the frequency sampling point, m = 0, 1, 2..., N-1, and Y[m] is the mth sampling output.

6. The speech detection method according to claim 5, characterized in that: In step 5, the values ​​of the Mel filter corresponding to speech signals of different frequencies are different. According to the formula And the following formula: Among them, mal represents the number of Mel filters, f() represents the interval frequency, and k represents the number of filters. According to the obtained Mel matrix H, the data Y[m] after Fourier transformation is filtered by the Mel filter. The data energy function E(m)=|Y(m)| 2 H.

7. The speech detection method according to claim 6, characterized in that: In step 6, the logarithmic operation is .

8. The speech detection method according to claim 7, characterized in that: In step 7, discrete cosine transform is used to obtain the corresponding Mel cepstral coefficients: n=1,2...,L (L is usually 12-16).

9. The speech detection method according to claim 1, characterized in that: In step 8, the algorithm considers the first 5 frames of data collected at startup to be all noise and stores this noise feature in the vector table as the initial noise data recorded as N1. The subsequent cache of the second 5 frames of data is set to N2 and considered to be the second noise data. According to T = N + k * (NN min ) method to determine the threshold value, where T is the threshold value, k is the coefficient, and different parameters are used to set different sensitivities, including 0, 1, 2, and 3, respectively. N is the current noise value, and N min is the minimum noise value.

10. The speech detection method according to claim 9, characterized in that: In step 8, after obtaining the Mel coefficient, when the Mel cepstral coefficient is much larger than T, it is confirmed as speech, and when no significant mutation occurs, it is considered as noise, and this is used as the N value to update the threshold T.

Images