A dynamic noise filtering intercom and a control method thereof

Abstract

The application discloses a dynamic noise filtering intercom and a control method thereof, and belongs to the technical field of wireless communication and voice signal processing. The intercom comprises a transient noise tracking suppressor and a dual-mode adaptive noise threshold encoder. The transient noise tracking suppressor is combined with a high-speed zero-crossing rate detection and a short-time energy ratio criterion to locate and suppress the impact pulse noise in the time domain in real time, and the response delay is lower than 2 ms. The dual-mode adaptive noise threshold encoder receives a net noise voice signal, carries out segmented sub-band power spectrum estimation on residual noise of the net noise voice signal, dynamically adjusts an ADPCM quantization step and an encoding noise threshold, and feeds back the quantization noise power spectrum to the suppressor, so that the suppressor adjusts the zero-crossing rate threshold and the energy ratio threshold according to the quantization noise power spectrum, and prevents the quantization noise from being misjudged as the transient noise. Through the closed-loop feedback mechanism, the application improves the voice intelligibility MOS value to above 4.2 in a high noise environment above 85 dB, and improves the signal-to-noise ratio by above 18 dB compared with a traditional noise reduction intercom.

Description

Technical Field

[0001] This invention relates to the field of wireless communication and voice signal processing technology, specifically to a dynamic noise filtering walkie-talkie and its control method. Background Technology

[0002] Walkie-talkies, as core terminal equipment for trunked radio communication, are widely used in high-noise working environments such as construction sites, mining operations, rail transportation, fire rescue, and power line inspection. In these environments, the voice signals collected by walkie-talkie microphones contain two distinctly different types of noise components: the first type is impulsive noise, including wind pulses, key collisions, tool knocking, and fabric rubbing sounds, characterized by sudden amplitude increases, short durations (typically 5ms to 50ms), and a significantly higher zero-crossing rate than normal speech; the second type is steady-state background noise, including motor operation sounds, engine roars, air conditioner compressor sounds, and environmental reverberation, characterized by relatively stable amplitudes, concentrated power spectrum distribution, and slow changes in the zero-crossing rate.

[0003] Existing noise reduction technologies for walkie-talkies mainly employ the following approaches: The first approach uses passive LC filters with fixed cutoff frequencies or active filters composed of operational amplifiers in the analog domain. These filters remove low-frequency noise components through high-pass or band-pass methods. However, the filtering characteristics of this type of approach are fixed and cannot distinguish between speech and impulse noise, resulting in the attenuation of high-frequency harmonic components of useful speech. The second approach uses DSP chips in the digital domain to perform FIR or IIR digital filtering. Noise suppression is achieved by setting fixed passband and stopband boundaries. While this approach offers better filtering accuracy than analog approaches, it lacks adaptability to changes in noise type. When impulse noise occurs in the environment, the fixed-parameter filter cannot quickly respond and suppress the impulse component while maintaining speech integrity. The third type of solution uses an environmental sensing module to collect environmental parameters such as temperature, humidity, and wind speed, and then switches the preset noise reduction mode or adjusts the filter coefficients accordingly. This type of solution relies on the indirect mapping relationship between environmental parameters and noise type. When impulsive noise and steady-state background noise occur simultaneously and their proportion changes dynamically, the environmental parameters cannot reflect the instantaneous changes in noise components, resulting in mode switching lag or unstable noise reduction effect.

[0004] None of the three solutions mentioned above resolves a key technical contradiction: suppressing impulse noise requires extremely low-latency time-domain detection and correction mechanisms, while suppressing steady-state background noise requires fine power spectrum estimation and adaptive filtering in the frequency domain. When these two types of processing are executed serially in the same signal path, the impulse noise suppression operation of the pre-processing stage alters the signal's energy distribution and spectral characteristics, thus distorting the noise floor estimate upon which the subsequent frequency-domain processing relies. Furthermore, the quantization noise introduced by the subsequent frequency-domain processing is misjudged as new impulse noise by the pre-processing time-domain processing, forming a positive feedback loop of false suppression. This contradiction results in existing walkie-talkies generally having a speech clarity MOS value below 3.5 in high-dynamic-noise environments (noise intensity above 85dB with alternating impulse and steady-state noise), failing to meet the stringent call quality requirements of emergency rescue and industrial dispatching. Summary of the Invention

[0005] The present invention aims to solve the above-mentioned technical problems and provides a dynamic noise filtering walkie-talkie and its control method. Through the coordinated operation and closed-loop feedback of a transient noise tracking suppressor and a dual-mode adaptive noise threshold encoder, the speech clarity MOS value is improved to above 4.2 in a high dynamic noise environment of above 85dB.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A dynamic noise filtering control method comprises a transient noise tracking suppressor and a dual-mode adaptive noise threshold encoder. The transient noise tracking suppressor performs high-speed zero-crossing rate detection and short-time energy ratio joint criterion on the raw audio signal acquired by the microphone, locating the start and end times of impulsive noise in the time domain and attenuating its amplitude to output a cleaned-noise speech signal. The dual-mode adaptive noise threshold encoder receives this cleaned-noise speech signal, performs segmented sub-band power spectrum estimation on its residual noise components, dynamically sets the coding quantization step size and coding noise threshold based on the estimation results, and simultaneously feeds back the quantized noise power spectrum characteristic value to the transient noise tracking suppressor. The transient noise tracking suppressor adjusts the zero-crossing rate threshold and short-time energy ratio threshold parameters based on this feedback value.

[0007] The core technical means of the transient noise tracking suppressor include: dividing the original audio signal into frames with a frame length of 0.5ms, with 8 sampling points per frame; calculating the positive zero-crossing rate and negative zero-crossing rate for each frame, and taking the larger value as the instantaneous zero-crossing rate index; when the instantaneous zero-crossing rate index exceeds the high threshold of 6 zero-crossing events per frame and the short-time energy ratio is lower than the low energy ratio threshold of 0.30, the current frame is determined to be in the impulse noise interval; for the continuous frame sequence determined to be in the noise interval, the rising edge slope and falling edge slope of its amplitude envelope are calculated; when the rising edge slope exceeds the rising edge slope threshold and the falling edge slope is lower than the falling edge slope threshold, an exponential decay amplitude correction is performed on the sampling points in the interval, with the correction coefficient increasing sequentially from 0.15 to 0.60. The above parameters are set based on the following: Through statistical analysis of 1,000 sets of field-measured impulse noise samples, the high threshold of 6 zero-crossing events per frame achieved a capture rate of 97.3% for wind impulse noise and a misjudgment rate of less than 0.8% for normal speech; the energy ratio threshold of 0.30 can effectively distinguish between the energy rise at the beginning of speech and the energy surge of impulse noise, and the short-time energy ratios of the two have obvious statistical differences.

[0008] The core technical means of the dual-mode adaptive noise threshold encoder include: dividing the noise-reduced speech signal into 32 sub-bands, each with a bandwidth of 312.5Hz, covering the entire frequency band from 0 to 10kHz; calculating the instantaneous power value of the current frame for each sub-band using a first-order recursive averaging algorithm, with a recursion coefficient of 0.875; smoothing and averaging the instantaneous power values ​​of 8 consecutive frames to obtain the sub-band noise floor estimate; calculating the ratio of the instantaneous power value to the noise floor estimate of each sub-band as the sub-band signal-to-noise ratio index; and setting the encoding... The noise threshold is the weighted average of the signal-to-noise ratio (SNR) of each sub-band. The weighting coefficient is positively correlated with the human voice sensitivity factor of that sub-band. The human voice sensitivity factor is 1.0 in the 300Hz to 3400Hz frequency band and 0.3 outside this frequency band. The quantization step size of the ADPCM encoder is dynamically adjusted according to the coding noise threshold. When the coding noise threshold is below 12dB, the quantization step size is reduced to 0.5 times the current value. When the coding noise threshold is above 24dB, the quantization step size is increased to 1.5 times the current value.

[0009] The core technical means of the closed-loop feedback mechanism include: the encoder subtracts the reconstructed signal from the original noise-reduced speech signal point by point to obtain the quantization error sequence; a 256-point FFT transform is performed on the quantization error sequence to obtain the quantization noise power spectrum; the average power density value in the 1kHz to 4kHz frequency band of the quantization noise power spectrum is extracted as the feedback feature value; the transient noise tracking suppressor adjusts the high threshold of the zero-crossing rate detection according to the feedback feature value, and the adjustment formula is that the high threshold update value is equal to the basic threshold plus the product of the feedback feature value and the scaling factor 0.02; at the same time, the short-time energy ratio threshold is adjusted according to the feedback feature value, and the adjustment formula is that the energy ratio threshold update value is equal to the basic threshold minus the product of the feedback feature value and the scaling factor 0.015. The working principle of this feedback mechanism is as follows: when the encoder quantization noise increases, the average power density in the 1kHz to 4kHz frequency band of the quantization noise power spectrum increases. Based on this, the transient noise tracking suppressor raises the zero-crossing rate threshold and lowers the energy ratio threshold, so that the suppressor no longer misjudges low zero-crossing rate components in the signal that are similar to quantization noise as impulsive impulse noise, thereby avoiding secondary attenuation of quantization noise; when the quantization noise decreases, the suppressor automatically restores the original threshold to maintain sensitive capture of real impulsive impulse noise.

[0010] The transient noise tracking suppressor performs tail smoothing after the exponential decay amplitude correction is completed: for the last three frames of data in the correction interval, the corrected amplitude value and the original amplitude value before correction are weighted and fused according to a linearly gradual ratio. The fusion ratio for the first frame is 0.75 for the corrected amplitude and 0.25 for the original amplitude; for the second frame, the weights are 0.50 for each; and for the third frame, the weights are 0.25 for the corrected amplitude and 0.75 for the original amplitude. The purpose of tail smoothing is to eliminate the abrupt amplitude change at the end of the exponential decay correction and prevent perceptible boundary effects between the end of the correction interval and the subsequent uncorrected signal. These boundary effects are audibly manifested as clicking or popping sounds.

[0011] The dual-mode adaptive noise threshold encoder dynamically adjusts the ADPCM quantization step size while simultaneously adjusting the walkie-talkie's RF transmit power. A linear mapping relationship is established between the quantization step size and the RF transmit power, with a mapping slope of 0.15 dBm per quantization step size unit. When the quantization step size decreases, the RF transmit power decreases synchronously according to the mapping relationship, with a decrease not exceeding 3 dBm; when the quantization step size increases, the RF transmit power increases synchronously, with an increase not exceeding 5 dBm. The technical effect of this linkage adjustment is as follows: a smaller quantization step size means the encoder represents the signal amplitude with finer granularity, resulting in reduced quantization distortion and improved signal quality. Therefore, the RF transmit power can be appropriately reduced to save battery power without affecting overall communication quality. Conversely, a larger quantization step size means the encoder represents the signal amplitude with coarser granularity, increasing quantization distortion. In this case, the RF transmit power needs to be appropriately increased to compensate for the decrease in signal-to-noise ratio caused by quantization distortion, ensuring that the receiver can correctly decode voice information. The RF transmit power is adjusted within 1ms after each encoded frame is completed, with an adjustment step of 0.5dBm each time. This response speed ensures that the RF power adjustment is synchronized with the encoding parameter adjustment, and there will be no mismatch where the encoder has switched the quantization step size but the RF power remains in the previous state.

[0012] The ADPCM quantization step size is adjusted every four encoded frames. The re-evaluation is based on the average coded noise threshold of the most recent four frames: if the average coded noise threshold is below 12dB, the quantization step size is reduced to 0.5 times the current value; if the average coded noise threshold is above 24dB, the quantization step size is increased to 1.5 times the current value; and if the average coded noise threshold is between 12dB and 24dB, the quantization step size remains unchanged. This four-frame re-evaluation cycle is designed because: an excessively short evaluation cycle (single-frame evaluation) causes the quantization step size to switch frequently during the alternation of voiced and unvoiced sounds in the speech signal, resulting in perceptible jitter in the encoded quality; an excessively long evaluation cycle (more than eight frames) causes the encoder to lag in response to sudden changes in noise, meaning the user can still hear the degraded speech for 40ms to 60ms after a noise spike. The response latency corresponding to the four-frame re-evaluation cycle is 20ms, which is within the range of the short-term masking effect of the human ear, so the user will not perceive the transition in encoded quality.

[0013] The walkie-talkie also includes an RF transceiver module, a baseband processing module, and a power management module. An adjustable attenuator is connected in series in the RF transceiver module's transmit link. The attenuation of this attenuator is controlled by the quantization step size signal of the dual-mode adaptive noise threshold encoder, with an attenuation adjustment range of 0dB to 8dB and an adjustment step of 1dB. The adjustable attenuator provides a second transmit power control path besides adjusting the RF power amplifier's supply voltage. When the quantization step size signal indicates a need to reduce transmit power, the attenuator increases the attenuation without changing the power amplifier's operating point, thus avoiding nonlinear distortion caused by frequent switching of bias voltage. The baseband processing module has a built-in hardware watchdog timer independent of the main control chip. When the frame processing timeout of either the transient noise tracking suppressor or the dual-mode adaptive noise threshold encoder exceeds 5ms, the hardware watchdog timer triggers a reset signal to restore the module to its initial state. This mechanism ensures that the digital signal processing link can automatically recover under extreme electromagnetic interference or software anomalies, preventing the walkie-talkie from being in a state of no audio output for an extended period due to a single processing anomaly.

[0014] Beneficial effects

[0015] The synergistic operation of the transient noise tracking suppressor and the dual-mode adaptive noise threshold encoder produces the following mutually reinforcing technical effects: First, the transient noise tracking suppressor eliminates over 90% of the impulsive noise components in the original audio signal, ensuring that the input signal for subband power spectrum estimation by the dual-mode adaptive noise threshold encoder does not contain large-amplitude pulse spikes. The fluctuation range of the subband noise floor estimate is reduced from ±6dB in the traditional scheme to ±1.5dB, and the setting accuracy of the coding noise threshold is improved by approximately 75%. The accurate coding noise threshold ensures that the ADPCM quantization step size is always within the optimal range, and the quantization signal-to-noise ratio does not experience an instantaneous degradation of more than 3dB when impulsive noise occurs.

[0016] Second, the quantization noise power spectrum characteristics fed back by the dual-mode adaptive noise threshold encoder enable the transient noise tracking suppressor to distinguish between real-world impulsive noise and quantization noise introduced by the encoder. The suppressor's misclassification rate for high-frequency unvoiced components in normal speech is reduced from 3.2% without feedback to below 0.8%. This reduction in misclassification rate directly reduces the unexpected attenuation of high-frequency harmonics in speech, improving the naturalness and intelligibility of speech by 12% and 8%, respectively.

[0017] Third, the linkage adjustment mechanism between the quantization step size and the radio frequency transmission power enables the walkie-talkie to automatically reduce the transmission power when the quantization step size decreases, extending the standby time by about 22% under the same battery capacity conditions; when the quantization step size increases, it automatically increases the transmission power, ensuring that the bit error rate at the receiving end does not increase due to the increase in quantization distortion. In an 85dB noise environment, the bit error rate of the receiving end voice frame is stable at less than 0.5%, while the bit error rate of traditional noise-canceling walkie-talkies in the same environment is about 2.8% to 4.5%.

[0018] Fourth, the voice sensitivity weighting mechanism in segmented subband power spectrum estimation tilts the setting weight of the coding noise threshold towards the 300Hz to 3400Hz frequency band, so that the encoder allocates more coding bits to the frequency band that contributes the most to speech intelligibility without changing the total bit rate. The formant structure of speech remains intact, and the speech intelligibility MOS value is improved by 0.4 to 0.6 compared with the uniform subband weighting scheme. Attached Figure Description

[0019] Figure 1 This is a block diagram of the overall structure of the dynamic noise filtering walkie-talkie of the present invention; Figure 2 Here is a flowchart of the transient noise tracking suppressor's operation. Figure 3 This is a flowchart of the operation of a dual-mode adaptive noise threshold encoder. Figure 4 This is a schematic diagram of the signal flow for a closed-loop feedback mechanism. Figure 5 A schematic diagram of frame weight allocation for tail smoothing; Figure 6 This is a graph showing the mapping relationship between quantization step size and RF transmit power. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that the following embodiments are merely illustrative of the technical solutions of the present invention and do not constitute any limitation on the scope of protection of the present invention.

[0021] Implementation Method 1

[0022] Reference Figure 1 A dynamic noise-filtering walkie-talkie includes a microphone array, an analog-to-digital converter, a transient noise tracking suppressor, a dual-mode adaptive noise threshold encoder, a baseband processing module, an RF transceiver module, a power amplifier, an antenna, a speaker, a digital-to-analog converter, a power management module, and a hardware watchdog timer.

[0023] The microphone array consists of two electret microphones spaced 15mm apart, connected differentially to the two input channels of the analog-to-digital converter (ADC). The ADC uses 16-bit precision and a sampling rate of 48kHz. The quantized digital audio signal is transmitted to the transient noise tracking suppressor via a serial audio interface.

[0024] Reference Figure 2 The transient noise tracking suppressor works as follows: Step S201: The digital audio signal is divided into frames with a frame length of 0.5ms, and each frame contains 24 sampling points (24 points in 0.5ms at a sampling rate of 48kHz). The framing is performed using a 50% overlapping Hanning window, and the overlap between adjacent frames is 12 sampling points.

[0025] Step S202: Calculate the positive and negative zero-crossing rates for each frame. The positive zero-crossing rate is calculated as follows: traverse adjacent sampling point pairs within the frame. If the preceding sampling point is negative or zero and the following sampling point is positive, increment the positive zero-crossing count by 1. After traversal, the total positive zero-crossing count is the positive zero-crossing rate. The negative zero-crossing rate is calculated as follows: traverse adjacent sampling point pairs within the frame. If the preceding sampling point is positive or zero and the following sampling point is negative, increment the negative zero-crossing count by 1. After traversal, the total negative zero-crossing count is the negative zero-crossing rate. The larger of the positive and negative zero-crossing rates is taken as the instantaneous zero-crossing rate index Z_inst for that frame.

[0026] Step S203: Calculate the short-time energy ratio R_energy for each frame. The short-time energy ratio is defined as the ratio of the short-time energy of the current frame to the short-time energy of the previous frame, where the short-time energy is the sum of the squares of the amplitudes of all sampling points within the frame. The formula for calculating the short-time energy ratio R_energy is:

[0027] Where E[n] is the short-time energy of the nth frame, and E[n-1] is the short-time energy of the (n-1)th frame.

[0028] Step S204: When Z_inst is greater than or equal to 6 and R_energy is less than or equal to 0.30, the current frame is determined to be in the impulsive noise region. Frames that continuously meet this criterion are marked as a noise frame sequence, and the start frame number N_start and end frame number N_end of the noise frame sequence are recorded.

[0029] Step S205: Calculate the amplitude envelope for the noisy frame sequence. The amplitude envelope is extracted by taking the maximum absolute value of the 24 sampling points within the frame, and using this maximum value as the amplitude envelope value A[n] for that frame. Calculate the rising edge slope S_rise and falling edge slope S_fall of the amplitude envelope. The rising edge slope is defined as:

[0030] The slope of the falling edge is defined as:

[0031] Step S206: When S_rise is greater than 0.15 and S_fall is less than -0.10, the noise frame sequence is determined to conform to the morphological characteristics of impulsive noise, and exponential decay amplitude correction is performed. The correction coefficient is calculated as follows: Let the total number of frames in the noise frame sequence be L, and the correction coefficient alpha[k] of the k-th frame (k starts counting from 0) is calculated according to the following formula:

[0032] That is, the correction factor increases linearly from 0.15 in the first frame to 0.60 in the last frame. The corrected amplitude at each sampling point is the product of the original amplitude and the correction factor:

[0033] Step S207: Perform tail smoothing processing. For the last 3 frames of the noisy frame sequence (i.e., frame L-3, frame L-2, and frame L-1), the corrected amplitude values ​​and the original amplitude values ​​before correction are weighted and fused according to a linear gradient ratio.

[0034] The fusion formula for frame L-3 is:

[0035] The fusion formula for frame L-2 is:

[0036] The fusion formula for frame L-1 is:

[0037] The signal output after tail smoothing is the noise-reduced speech signal, which is transmitted to the dual-mode adaptive noise threshold encoder.

[0038] Reference Figure 3 The workflow of the dual-mode adaptive noise threshold encoder is as follows: Step S301: The noise-reduced speech signal is divided into frames with a frame length of 10ms and a frame shift of 5ms. Each frame contains 480 sampling points.

[0039] Step S302: Perform power spectrum division of 32 sub-bands for each frame. The sub-band division is implemented using a polyphase filter bank, with the center frequencies of each sub-band being 156.25Hz, 468.75Hz, 781.25Hz, and so on up to 9843.75Hz, and the bandwidth of each sub-band is 312.5Hz.

[0040] Step S303: For each sub-band, calculate the instantaneous power value P_sub[m][n] of the current frame using a first-order recursive averaging algorithm, where m is the sub-band index (0 to 31) and n is the frame index. The recursive calculation formula is:

[0041] Where X_sub[m][n] are the complex spectral coefficients of the m-th subband in the n-th frame.

[0042] Step S304: Smoothly average the instantaneous power values ​​of 8 consecutive frames to obtain the sub-band noise basis estimate N_floor[m][n]:

[0043] Step S305, calculate the subband signal-to-noise ratio (SNR) index SNR_sub[m][n] for each subband:

[0044] Step S306: Set the human voice sensitivity weight vector W[m], where m ranges from 0 to 31. The sub-band indices corresponding to 300Hz to 3400Hz are m=1 to m=10 (156.25Hz multiplied by 2 to 156.25Hz multiplied by 22), and the weight of these sub-bands is 1.0; the weight of the remaining sub-bands is 0.3. Calculate the coded noise threshold T_noise:

[0045] Step S307: Adjust the ADPCM quantization step size delta according to the coding noise threshold T_noise. A re-evaluation is performed every four coded frames. Let the average coding noise threshold of the most recent four frames be T_avg. When T_avg is below 12dB, delta is updated to 0.5 times the current value; when T_avg is above 24dB, delta is updated to 1.5 times the current value; when T_avg is between 12dB and 24dB, delta remains unchanged.

[0046] Step S308: Synchronously adjust the RF transmit power P_tx according to the quantization step size delta. The linear mapping relationship between the quantization step size delta and the RF transmit power P_tx is as follows:

[0047] Where P_tx0 is the reference transmit power and delta_0 is the reference quantization step size. When delta decreases, P_tx decreases, with a decrease not exceeding 3dBm; when delta increases, P_tx increases, with an increase not exceeding 5dBm. The adjustment is performed within 1ms after each encoded frame is completed, with an adjustment step of 0.5dBm each time. The adjustable attenuator connected in series in the RF transceiver module transmit link synchronously adjusts the attenuation amount. When P_tx decreases, the attenuation amount increases from the current value by 1dB; when P_tx increases, the attenuation amount decreases from the current value by 1dB. The attenuation adjustment range is 0dB to 8dB.

[0048] Reference Figure 4 The signal flow of the closed-loop feedback mechanism is as follows: After encoding each frame, the dual-mode adaptive noise threshold encoder calculates the difference between the reconstructed signal and the original noise-reduced speech signal point by point to obtain the quantization error sequence e[i], where i ranges from 0 to 479 (480 sampling points per frame). A 256-point FFT is performed on the quantization error sequence. Since the frame length is 480 points, the first 256 points are truncated for the transformation. The quantization noise power spectrum E_fft[k] is obtained after the FFT, where k ranges from 0 to 127 (the first 128 points of the symmetric spectrum are used).

[0049] Extract the average power density value F_fb from the quantization noise power spectrum in the frequency band from 1kHz to 4kHz. 1kHz to 4kHz corresponds to FFT indices k=5 to k=21 (frequency resolution 48kHz / 256=187.5Hz). The formula for calculating the average power density value is:

[0050] F_fb is transmitted as a feedback feature value to the transient noise tracking suppressor. Upon receiving F_fb, the transient noise tracking suppressor updates the zero-crossing rate high threshold Z_th and the short-time energy ratio threshold R_th:

[0051]

[0052] Where Z_th0 is the base value of 6 for the zero-crossing rate high threshold, and R_th0 is the base value of 0.30 for the short-time energy ratio threshold. When F_fb increases, Z_th increases and R_th decreases, making the noise judgment condition of the suppressor more stringent and reducing the probability of misjudging quantization noise; when F_fb decreases, Z_th decreases and R_th increases, and the suppressor regains its sensitive capture of impulsive noise.

[0053] Implementation Method 2

[0054] The difference between this implementation method and Implementation Method 1 is that the frame length of the transient noise tracking suppressor is adjusted to 1.0 ms, the number of sampling points per frame is 48, and the inter-frame overlap rate is adjusted to 75%. The purpose of adjusting the frame length and overlap rate is to adapt to noise environments dominated by mechanical collision noise (such as the operating noise of mining equipment). In mechanical collision noise environments, the duration of impact impulse noise is typically 20 ms to 80 ms, longer than wind impulse noise. Using a 1.0 ms frame length reduces the number of frame processing operations, lowering the processor's computational load. The 75% overlap rate ensures accurate detection of the noise start and end times even with a longer frame length, achieving detection accuracy comparable to the 0.5 ms frame length with a 50% overlap rate scheme in Implementation Method 1.

[0055] The high threshold of the transient noise tracking suppressor is adjusted to 10 zero-crossing events per frame, and the low energy ratio threshold is adjusted to 0.25. The zero-crossing rate characteristics of mechanical collision noise differ from those of wind impulse noise. Its spectral energy is concentrated in the low-frequency band of 200Hz to 800Hz, with a relatively low zero-crossing rate but a larger energy surge. Therefore, it is necessary to reduce the energy ratio threshold to improve the sensitivity to energy surge characteristics.

[0056] The voice sensitivity weights of the dual-mode adaptive noise threshold encoder are adjusted as follows: 1.0 for the 300Hz to 3400Hz band, 0.5 for the 3400Hz to 8000Hz band, and 0.2 for the 8000Hz to 10000Hz band. In mining environments, the energy of high-frequency noise (such as drilling noise and conveyor belt friction noise) is concentrated in the 3000Hz to 6000Hz band. Appropriately increasing the voice sensitivity weight in this band allows the encoder to assign greater weight to high-frequency speech harmonics in subband signal-to-noise ratio estimation, preventing the encoder from over-compressing high-frequency speech components due to high-frequency noise interference.

[0057] The mapping slope between quantization step size and RF transmit power is adjusted to 0.20 dBm per quantization step size unit. In mining environments, walkie-talkie communication distances are typically short (200m to 800m), and the RF link margin is ample. Therefore, the power adjustment range can be appropriately increased to more fully utilize the energy-saving potential brought about by changes in quantization step size.

[0058] Implementation Method 3

[0059] The difference between this implementation and implementation one is that the subband division method of the dual-mode adaptive noise threshold encoder is changed from a polyphase filter bank to a uniform subband division based on FFT. The number of FFT points is set to 512, and the frequency resolution corresponding to a sampling rate of 48kHz is 93.75Hz. Every 4 FFT frequency points are combined into one subband, resulting in a total of 128 subbands, each with a bandwidth of 375Hz. The purpose of increasing the number of subbands from 32 to 128 is to improve the frequency resolution of the subband signal-to-noise ratio estimation, enabling the encoder to more finely distinguish between noise frequency points and speech frequency points.

[0060] The calculation method for the coding noise threshold has been adjusted as follows: The 128 sub-bands are divided into three groups according to human voice sensitivity. The first group consists of sub-bands from 200Hz to 4000Hz (sub-band indices 3 to 42), with a weight of 1.0; the second group consists of sub-bands from 4000Hz to 7000Hz (sub-band indices 43 to 74), with a weight of 0.6; and the third group consists of sub-bands below 200Hz and above 7000Hz, with a weight of 0.2. The coding noise threshold is the weighted average of the signal-to-noise ratio (SNR) of the three sub-band groups.

[0061] The quantization step size adjustment cycle is shortened to a re-evaluation every two coded frames. The higher frequency resolution of the uniform subband division in FFT results in a more sensitive subband signal-to-noise ratio estimate to noise changes, allowing for a shorter evaluation cycle without frequent jitter in the quantization step size. The response latency corresponding to a two-frame evaluation cycle is 10ms, half that of 20ms in Implementation Method 1, enabling the encoder to track changes in the noise environment more quickly.

[0062] Implementation Method 4

[0063] The difference between this implementation method and Implementation Method 1 is that the frequency range for extracting the quantization noise power spectrum characteristic value F_fb output by the dual-mode adaptive noise threshold encoder is expanded from 1kHz to 4kHz to 500Hz to 6kHz. The basis for expanding the extraction range is that when the encoder operates at a lower bit rate (larger ADPCM quantization step size), the power spectrum distribution of the quantization noise expands towards lower frequencies, and the extraction range of 500Hz to 6kHz can more comprehensively reflect the energy distribution characteristics of the quantization noise.

[0064] The scaling factor in the threshold update formula for the transient noise tracking suppressor is adjusted as follows: the scaling factor for the zero-crossing rate high threshold is 0.025, and the scaling factor for the short-time energy ratio threshold is 0.020. Increasing the scaling factor increases the threshold's response to changes in quantization noise, allowing the suppressor to be adjusted to a suitable operating point more quickly when quantization noise fluctuates drastically. This adjustment is suitable for use of walkie-talkies in environments with severe electromagnetic interference, which can cause the ADPCM encoder to frequently switch quantization step sizes, leading to rapid fluctuations in quantization noise power.

[0065] The number of frames for smoothing the tail has been increased from 3 to 5. The purpose of increasing the number of frames is to further smooth the amplitude transition at the end of the correction interval when the quantization step size changes frequently, preventing the accumulation of boundary effects caused by changes in quantization noise. The fusion weights for the 5 frames are as follows: frame L-5, the corrected amplitude weight is 0.85 and the original amplitude weight is 0.15; frame L-4, the weights are 0.70 and 0.30; frame L-3, the weights are 0.50 and 0.50; frame L-2, the weights are 0.30 and 0.70; and frame L-1, the weights are 0.15 and 0.85.

[0066] Implementation Method 5

[0067] The difference between this implementation and implementation one is that the adjustment range of the adjustable attenuator in the RF transceiver module's transmit link is expanded from 0dB to 8dB to 0dB to 16dB, with the adjustment step remaining at 1dB. After expanding the adjustment range, the walkie-talkie can reduce its transmit power to a lower level when the quantization step size is reduced to its minimum value (i.e., when the encoding accuracy is highest), further saving battery power. The control signal for the adjustable attenuator is generated by mapping the quantization step size signal of the dual-mode adaptive noise threshold encoder through a first-level lookup table. The lookup table stores the attenuation values ​​corresponding to 128 quantization step sizes, with a mapping accuracy of 0.5dB per level.

[0068] The power management module features dynamic voltage regulation, dynamically adjusting the digital core supply voltage based on real-time load calculations from the transient noise tracking suppressor and the dual-mode adaptive noise threshold encoder. When both modules are under low load (frame processing time less than 60% of the frame length), the supply voltage decreases by 0.15V from its nominal value; when either module is under high load (frame processing time exceeding 85% of the frame length), the supply voltage returns to its nominal value. This dynamic voltage regulation, combined with the synchronized adjustment of RF transmit power, reduces the overall power consumption of the walkie-talkie in quiet environments by approximately 18% compared to a fixed voltage solution.

[0069] The above description is merely a specific embodiment of the present invention and does not limit the scope of patent protection of the present invention. Any equivalent structural or procedural modifications made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A dynamic noise filtering walkie-talkie control method, characterized in that, It consists of a transient noise tracking suppressor and a dual-mode adaptive noise threshold encoder, where the transient noise tracking suppressor and the dual-mode adaptive noise threshold encoder are independent hardware logic modules. The transient noise tracking suppressor performs high-speed zero-crossing rate detection and short-time energy ratio joint judgment on the raw audio signal collected by the microphone, locates the start and end times of the impulsive impulse noise in the time domain and attenuates its amplitude to output a clean speech signal. The dual-mode adaptive noise threshold encoder receives the clean speech signal, performs segmented sub-band power spectrum estimation on its residual noise components, dynamically sets the coding quantization step size and coding noise threshold based on the estimation results, and feeds back the quantization noise power spectrum feature value to the transient noise tracking suppressor. The transient noise tracking suppressor adjusts the zero-crossing rate threshold and short-time energy ratio threshold parameters based on this feedback value, so that the suppression depth of the transient noise tracking suppressor for impulsive noise is matched with the quantization noise floor of the dual-mode adaptive noise threshold encoder.

2. The dynamic noise filtering walkie-talkie control method according to claim 1, characterized in that, The method for high-speed zero-crossing rate detection by the transient noise tracking suppressor includes: dividing the original audio signal into frames with a frame length of 0.5ms, with 8 sampling points per frame; calculating the positive and negative zero-crossing rates for each frame, where the positive zero-crossing rate is the number of times the signal crosses from a negative value to a positive value, and the negative zero-crossing rate is the number of times the signal crosses from a positive value to a negative value; taking the larger of the positive and negative zero-crossing rates as the instantaneous zero-crossing rate index for that frame; when the instantaneous zero-crossing rate index exceeds a preset high threshold and the short-time energy ratio is lower than a preset low energy ratio threshold, determining that the current frame is in the impulse noise interval; for the continuous frame sequence determined to be in the noise interval, calculating the rising slope and falling slope of its amplitude envelope; when the rising slope exceeds the rising slope threshold and the falling slope is lower than the falling slope threshold, performing exponential decay amplitude correction on the sampling points in that interval, with the correction coefficient increasing sequentially from 0.15 to 0.

60.

3. The dynamic noise filtering walkie-talkie control method according to claim 1, characterized in that, The method for segmented subband power spectrum estimation by a dual-mode adaptive noise threshold encoder includes: dividing the net-noise speech signal into 32 subbands, each with a bandwidth of 312.5 Hz, covering the entire frequency band from 0 to 10 kHz; calculating the instantaneous power value of the current frame for each subband using a first-order recursive averaging algorithm, with a recursion coefficient of 0.875; smoothing and averaging the instantaneous power values ​​of 8 consecutive frames to obtain the subband noise floor estimate; and calculating the ratio of the instantaneous power value to the noise floor estimate of each subband as the subband signal-to-noise ratio index. The coding noise threshold is set as the weighted average of the signal-to-noise ratio (SNR) of each sub-band. The weighting coefficient is positively correlated with the human voice sensitivity factor of that sub-band. The human voice sensitivity factor is 1.0 in the 300Hz to 3400Hz frequency band and 0.3 outside this frequency band. The quantization step size of the ADPCM encoder is dynamically adjusted according to the coding noise threshold. When the coding noise threshold is below 12dB, the quantization step size is reduced to 0.5 times the current value. When the coding noise threshold is above 24dB, the quantization step size is increased to 1.5 times the current value.

4. The dynamic noise filtering walkie-talkie control method according to claim 1, characterized in that, The method of feeding back the quantization noise power spectrum feature value to the transient noise tracking suppressor in a dual-mode adaptive noise threshold encoder includes: the encoder subtracts the original noise-reduced speech signal from the reconstructed signal of each frame bit by bit to obtain a quantization error sequence; a 256-point FFT transform is performed on the quantization error sequence to obtain the quantization noise power spectrum; the average power density value in the 1kHz to 4kHz frequency band of the quantization noise power spectrum is extracted as the feedback feature value; the transient noise tracking suppressor adjusts the high threshold of the zero-crossing rate detection according to the feedback feature value, and the adjustment formula is that the updated value of the high threshold is equal to the basic threshold plus the product of the feedback feature value and the scaling factor, and the scaling factor is 0.02; at the same time, the short-time energy ratio threshold is adjusted according to the feedback feature value, and the adjustment formula is that the updated value of the energy ratio threshold is equal to the basic threshold minus the product of the feedback feature value and the scaling factor, and the scaling factor is 0.

015.

5. The dynamic noise filtering walkie-talkie control method according to claim 1, characterized in that, The dual-mode adaptive noise threshold encoder dynamically adjusts the ADPCM quantization step size while simultaneously adjusting the radio frequency (RF) transmit power of the walkie-talkie: a linear mapping relationship is established between the quantization step size and the RF transmit power, with a mapping slope of 0.15 dBm per quantization step size unit; when the quantization step size decreases, the RF transmit power decreases synchronously according to the mapping relationship, with a decrease not exceeding 3 dBm; when the quantization step size increases, the RF transmit power increases synchronously, with an increase not exceeding 5 dBm; the adjustment of the RF transmit power is performed within 1 ms after each encoded frame is completed, with an adjustment step of 0.5 dBm each time.

6. The dynamic noise filtering walkie-talkie control method according to claim 1, characterized in that, The high threshold of the transient noise tracking suppressor is set to 6 zero-crossing events per frame, and the low energy ratio threshold is set to 0.

30. The high threshold of 6 zero-crossing events per frame was obtained by statistical analysis of 1,000 sets of field-measured impulse noise samples. Under this threshold, the capture rate of wind impulse noise reaches 97.3% and the misjudgment rate of normal speech is less than 0.8%.

7. The dynamic noise filtering walkie-talkie control method according to claim 1, characterized in that, The ADPCM quantization step size adjustment cycle of the dual-mode adaptive noise threshold encoder is to perform a re-evaluation every 4 coded frames. The re-evaluation is based on the average coded noise threshold of the most recent 4 frames. When the average coded noise threshold is below 12dB, the quantization step size is reduced to 0.5 times the current value. When the average coded noise threshold is above 24dB, the quantization step size is increased to 1.5 times the current value. When the average coded noise threshold is between 12dB and 24dB, the quantization step size remains unchanged. This 4-frame re-evaluation cycle enables the encoder to keep the response delay within 20ms when there are sudden changes in the noise environment.

8. The dynamic noise filtering walkie-talkie control method according to claim 2, characterized in that, After the exponentially decaying amplitude correction is completed, the transient noise tracking suppressor performs tail smoothing: for the last 3 frames of data in the correction interval, the corrected amplitude value and the original amplitude value before correction are weighted and fused according to a linearly gradual ratio. The fusion ratio of the first frame is 0.75 for the corrected amplitude and 0.25 for the original amplitude, the weight of the second frame is 0.50 for each, and the weight of the third frame is 0.25 for the corrected amplitude and 0.75 for the original amplitude. After tail smoothing, the final noise-reduced speech signal is output.

9. The control method for a dynamic noise filtering walkie-talkie according to claim 1, characterized in that, After the dual-mode adaptive noise threshold encoder outputs the quantized noise power spectrum characteristic value, the transient noise tracking suppressor adjusts the zero-crossing rate threshold and short-time energy ratio threshold parameters based on this characteristic value. An adjustable attenuator is connected in series in the transmit link of the RF transceiver module. The attenuation of this adjustable attenuator is controlled by the quantization step size signal of the dual-mode adaptive noise threshold encoder. The attenuation adjustment range is 0dB to 8dB, and the adjustment step is 1dB. The baseband processing module has a built-in hardware watchdog timer independent of the main control chip. When the frame processing timeout of either the transient noise tracking suppressor or the dual-mode adaptive noise threshold encoder exceeds 5ms, the hardware watchdog timer triggers a reset signal to restore the module to its initial state.

10. A dynamic noise-filtering walkie-talkie, characterized in that, Perform the dynamic noise filtering walkie-talkie control method as described in any one of claims 1 to 9.

Images