Radio station AM voice signal quality evaluation method based on adaptive threshold
By adopting adaptive threshold and signal-to-noise ratio evaluation methods in radio AM speech signal quality evaluation, combining power spectrum and short-term energy-entropy ratio, the accuracy and computational complexity of signal quality evaluation in complex electromagnetic environments is solved, and the problem of Fast and robust signal quality assessment.
Patent Information
- Application Number
- CN202411277471.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-12-06
AI Technical Summary
The prior art is difficult to accurately evaluate signal quality in complex electromagnetic environments, and the calculation complexity is high, so the evaluation results cannot be quickly given, which affects the adjustment of signal processing.
Using the radio AM speech signal quality evaluation method based on adaptive threshold, the signal-to-noise ratio is used as the evaluation condition, and the signal-to-noise ratio is roughly estimated and corrected by combining the power spectrum and short-term energy-entropy ratio to enhance the robustness and fast algorithm Assessment ability.
It improves the accuracy of signal quality evaluation in complex electromagnetic environments, reduces the calculation complexity and hardware resource consumption, and quickly gives the received signal quality evaluation results, which facilitates subsequent signal processing.
Smart Images

Figure CN119091927A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of quality assessment, and further relates to a radio station voice signal quality assessment method, which can be used for shipborne long-distance radio station communication. Background Art
[0002] Evaluating the voice signal quality of the received signal is an important part of long-distance wireless communication signal processing. By timely and accurately judging the quality of the received voice signal, the relevant parameters of the signal transmitter and the processing method of the received signal can be adjusted in time, thereby improving the signal quality of the received signal and laying the foundation for subsequent signal processing. In recent years, researchers have proposed many solutions for the technology of voice signal quality evaluation.
[0003] The patent document with application number CN202111154709.3 discloses a "signal quality assessment device and signal quality assessment method" in the field of communication. Its signal quality assessment method is: the sender and the receiver agree in advance that both parties use the same pseudo-random code. Before each task, the sending station sends the pseudo-random code signal repeatedly and continuously after the pseudo-random code signal is powered on. After the pseudo-random code signal is sent, the synchronization header and the formal message signal are sent. After receiving the signal, the receiving station amplifies and demodulates the low-frequency signals such as random code signals and noise, and performs correlation calculations on the demodulated signal and the pre-stored pseudo-random code signal to determine the amplitude of the pseudo-random code signal in the low-frequency signal; according to the amplitude of the low-frequency signal and the amplitude of the pseudo-random code signal, the square difference between the amplitude of the low-frequency signal and the amplitude of the pseudo-random code signal is calculated to obtain the noise variance; the signal-to-noise ratio is determined according to the noise variance and the power of the pseudo-random code signal, and the communication signal quality is evaluated according to the range of the estimated result of the signal-to-noise ratio. Although this method can estimate the signal-to-noise ratio at a relatively fast speed and evaluate the signal quality through the signal-to-noise ratio estimation result, which is convenient for hardware device implementation, it is necessary to transmit a pseudo-random coded signal before transmitting the signal to provide a reference signal when estimating the signal-to-noise ratio. Therefore, in a complex electromagnetic environment, the pseudo-random coded signal is easily affected by the interference signal, which makes it difficult to estimate the power of the reference signal and makes it impossible to implement signal quality evaluation in a complex electromagnetic environment with strong interference.
[0004] The patent document with application number CN201710475912.8 discloses "an output-based objective speech quality evaluation method". The specific steps of the method are: calculating the Mel-frequency cepstral coefficients of the distorted speech after system transmission; constructing a reference model according to the auditory characteristics of the human ear; calculating the consistency measurement of the Mel-frequency cepstral coefficients of the distorted speech and the reference model that conforms to the auditory characteristics of the human ear; inserting a sequence into the original speech, and calculating the bit error rate of the sequence extracted from the distorted speech after system transmission; establishing a mapping relationship between the subjective MOS score and the consistency measurement according to the consistency measurement and the bit error rate, and obtaining an objective prediction model for the MOS score of the speech to be evaluated, and objectively evaluating the speech quality through the objective prediction model. Since this method evaluates the signal quality only by calculating the consistency measurement of the Mel-frequency cepstral coefficients and the reference model that conforms to the auditory characteristics of the human ear, it cannot directly reflect the influence of the interference intensity in the received signal on the signal quality under complex interference conditions, which is not convenient for the subsequent signal processing method to be adjusted, and the calculation complexity is large during the evaluation, which is not conducive to the rapid presentation of the actual evaluation results. Summary of the invention
[0005] The purpose of the present invention is to address the defects of the above-mentioned prior art and propose a radio AM voice signal quality assessment method based on adaptive threshold, so as to broaden the algorithm's applicable electromagnetic environment, improve robustness, reduce the computational complexity during operation, and reduce hardware resource consumption.
[0006] To achieve the above purpose, the technical idea of this method is: select the signal-to-noise ratio as the signal quality evaluation condition to avoid the high amount of calculation required for evaluation through an objective prediction model, and at the same time perform a rough estimation of the signal-to-interference-noise ratio through power spectrum cancellation and obtain the proportional relationship between the received signal with and without speech segments through adaptive threshold endpoint detection, and then correct the rough estimation result through the proportional relationship, thereby improving the accuracy of the signal-to-interference-noise ratio estimation result and enhancing the robustness of the algorithm.
[0007] According to the above ideas, the technical solutions of the present invention include the following:
[0008] (1) Segment the received signal X(n) to obtain the segmented received signal x i (n)n=1,2···,L1,i=1,2···,N1,N1 is the number of segments, L1 is the signal length after segmentation;
[0009] (2) Demodulate the received signal X(n) to obtain a speech signal A(n);
[0010] (3) Segment the demodulated speech signal A(n) to obtain segmented speech signal a i (n);
[0011] (4) For the segmented speech signal a i (n) Frame division and calculate the s-th frame signal b s (m) short-time energy entropy ratio Ef i (s), and average the short-time energy entropy ratio of all frame signals of this segment to obtain the segmented speech signal a i (n) average short-time energy entropy ratio Efi;
[0012] (5) Preset the retention value EfB, and compare the short-time energy entropy ratio of each frame signal to Ef i (s) is compared with it to obtain the preliminary presence or absence of the speech frame of the signal:
[0013] If the short-time energy entropy ratio of the signal of more than 5 consecutive frames is Ef i If (s) is greater than or equal to EfB*(1+δ), it is regarded as a preliminary continuous speech frame and step (6) is executed, wherein δ is a parameter for judging the speechless segment obtained by experiments.
[0014] Otherwise, it is considered that all the frames in the signal segment are silent frames;
[0015] (6) Calculate the average short-time energy entropy ratio EfA of the initial continuous frames with conversation and the final endpoint detection threshold EfI, and convert the short-time energy entropy ratio EfA of the frames with conversation into i (s) is compared with the threshold to obtain the final detection result:
[0016] If the short-time energy entropy ratio Efx(s) of more than 4 consecutive frames of signals is greater than or equal to EfI, the detected continuous talking frames are regarded as the final continuous talking frames, and EfB=EfB*λ2+(1-λ2)*EfI, where 0≤λ2≤1.
[0017] Otherwise, the initial continuous speech frames are all regarded as no-speech frames, and EfB=EfBλ3+(1-λ3)*EfI is updated, where 0≤λ3≤1, and λ2≤λ3;
[0018] (7) Repeat steps (5) to (6) until the detection of the speech signal of the current segment is completed, and calculate the ratio α of the speech frames with speech in the current segment of the speech signal to the total number of frames;
[0019] (8) Estimate the signal-to-interference-plus-noise ratio (ESIR) of this segment using the ratio α i :
[0020] (8a) Calculate the segmented received signal x i Power spectrum PF of (n) i (f r ), through the power spectrum PF i (f r ) Estimate the segment received signal x i (n) Carrier power: P zbi= PF i (f zb - 1) + PF i (f zb ) + PF i (f zb + 1), where f zb is the position corresponding to the carrier frequency point;
[0021] (8b) Use the ratio α to correct the estimated result P zbi of the carrier power to obtain the corrected power of the speech signal where, is the power ratio of the carrier signal to the speech signal in a frame of received signal with speech and no interference obtained from the experiment;
[0022] (8c) Estimate the signal-to-interference ratio of the segmented received signal: ESIR hy = 10 * log((P zbi + P i )) / (P zbi - P hy ), where P i is the total power of the segmented received signal x zb (n); i is the total power of the segmented received signal x i (n);
[0023] (9) Repeat steps (4) to (8) until the signal-to-interference ratio of all segmented received signals x i (n) is estimated, and the estimated results of the signal-to-interference ratios of all segmented received signals are ESIR1, ESIR2... ESIR i …, ESIR N1 ;
[0024] (10) Select the maximum value ESIR = max(ESIR1, ESIR2... ESIR i …, ESIR N1 ) among all the estimated results of the segmented signals as the estimated result of the signal-to-interference ratio of the current received signal, and judge the signal quality of the received signal according to the result:
[0025] If the estimated result ESIR > 6dB, it is regarded as interference-free;
[0026] If - 5dB < ESIR ≤ 6dB, it is regarded as weak interference;
[0027] If - 9dB < ESIR ≤ - 5dB, it is regarded as medium interference;
[0028] If ESIR ≤ - 9dB, it is regarded as strong interference.
[0029] The present invention has the following advantages:
[0030] First, since the present invention selects the signal-to-interference-noise ratio as the evaluation feature, it can more intuitively reflect the impact of the intensity of artificial interference on the signal quality, and the calculation amount is relatively small compared to building a human ear model, which improves the running speed of the hardware algorithm and can provide an evaluation result of the received signal quality relatively quickly, which is convenient for subsequent received signal processing.
[0031] Second, the present invention combines the power spectrum characteristics of the AM radio signal with the endpoint detection results of the demodulated voice signal to estimate the signal-to-interference-noise ratio, which can improve the accuracy of signal quality estimation in complex electromagnetic interference environments and the interference application range of the algorithm.
[0032] Thirdly, the present invention improves the accuracy of endpoint detection, reduces the prerequisites required for endpoint detection, and enhances the robustness of the overall algorithm by updating the adaptive threshold of endpoint detection in real time through different parameters according to different situations of the voice segment.
[0033] Fourthly, the present invention corrects the estimation result of the carrier signal power in combination with the distribution of voice in the communication signal to obtain the estimation result of the voice power, thereby improving the accuracy of the final signal-to-interference-noise ratio estimation and the accuracy of the received signal quality evaluation. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a general implementation flow chart of the present invention;
[0035] Figure 2 It is a sub-flow chart of adaptive endpoint detection in the present invention;
[0036] Figure 3 This is a diagram of estimation results of a simulation experiment of signal-to-interference-noise ratio estimation using the present invention under the NoiFM interference type;
[0037] Figure 4 The figure is an estimation result diagram of a simulation experiment of signal-to-interference-noise ratio estimation under QPSK interference type using the present invention;
[0038] Figure 5 The figure is an estimation result diagram of a simulation experiment of signal-to-interference-noise ratio estimation under 2FSK interference type using the present invention;
[0039] Figure 6 The figure is an estimation result diagram of a simulation experiment of signal-to-interference-and-noise ratio estimation under BPSK interference type using the present invention. DETAILED DESCRIPTION
[0040] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work should all fall within the scope of protection of the present invention.
[0041] It should be noted that the step numbers in the specification and claims of the present invention are only for a clear description of the implementation scheme of the present invention to facilitate understanding, and the order of the step numbers is not limited.
[0042] Reference Figure 1 , the implementation steps of this example are as follows:
[0043] Step 1: Determine whether there is a received signal.
[0044] Before estimating the signal-to-interference-to-noise ratio of the received signal, it is necessary to determine whether the required received signal is received. When it is determined that there is a need to receive a signal, the signal-to-interference-to-noise ratio estimation is performed. Otherwise, the signal quality is directly output as strong interference. The specific judgment is as follows:
[0045] (1.1) Calculate the power spectrum PFz(f m ):
[0046]
[0047] where f m =(N / 2:N / 21)*(Fs / N) is the power spectrum frequency of the current received signal X'(n), Fs is the sampling rate of the required received signal X(n), and N is the length of the received signal;
[0048] (1.2) For the power spectrum PFz(f m ) in the frequency range [f a ,f b ] and sum to get the bandwidth power P z , f a f is the starting frequency of the power spectrum within the power spectrum bandwidth of the received signal, b The frequency point at the end of the power spectrum within the power spectrum bandwidth of the signal to be received;
[0049] (1.3) Set detection parameters Calculate the judgment ratio B = P z / (sum(PFz(f m ))-P z ), compare the two parameters:
[0050] If B is greater than the detection parameter It is considered that the signal that needs to be received is received, and step 2 is executed.
[0051] Otherwise, the quality assessment result is directly output as strong interference.
[0052] Step 2: Segment the received signal.
[0053] The received signal X(n) is segmented according to the set signal length L1 in the time domain to obtain a segmented received signal:
[0054] x i (n) = X((i-1)*L1:i*L1)
[0055] Among them, x i (n) is the segmented received signal of the i-th segment, i=1 2···N1, N1 is the number of segments, and * indicates multiplication;
[0056] Step 3: Demodulate the received signal X(n) to obtain a demodulated speech signal A(n).
[0057] (3.1) The received signal X(n) and the carrier signal X zb (n) = cos(2π*fc*n) to obtain the multiplication signal A”(n):
[0058] A”(n)=X(n)*X zb (n)
[0059] Where fc is the carrier frequency of the received signal X(n);
[0060] (3.2) The multiplication signal A" (n) is low-pass filtered at a frequency of fd*2 / Fs to obtain the filtered signal A' (n), where fd is the carrier signal bandwidth;
[0061] (3.3) Remove the DC component mean(A(n)′) from the filtered signal A′(n) to obtain the demodulated speech signal A(n):
[0062] A(n)=A(n)′-mean(A(n)′).
[0063] Step 4: Segment the speech signal A(n).
[0064] The speech signal A(n) is segmented in the time domain according to the set signal length L1 to obtain the segmented speech signal:
[0065] a i (n) = A((i-1)*L1:i*L1)
[0066] Among them, a i(n) is the segmented speech signal of the i-th segment.
[0067] Step 5: Segment the speech signal a i (n) Frame and obtain the final frame signal b s (m).
[0068] (5.1) In the segmented speech signal a i (n) Overlapping frames are performed on the time domain length to obtain the preliminary frame signal bl s (m):
[0069] bl s (m) = a i ((s-1)*inc:(s-1)*inc+wlen)
[0070] Wherein, wlen=L1 / 50 is the frame length, inc=wlen / 2 is the frame shift, m=1, 2…wlen, s=1, 2…fn, fn is the number of sub-frames;
[0071] (5.2) Window all preliminary framed signals to obtain the final framed signal b s (m):
[0072] b s (m) = bl s (m)*w(m)
[0073] in is the Hamming window.
[0074] Step 6: Calculate all final frame signals b s The short-time energy-entropy ratio of (m) and the average value of its short-time energy-entropy ratio Efi.
[0075] (6.1) Calculate the frame signal b s The short-time logarithmic energy LE of (m) s :
[0076]
[0077] Where a is a preset constant;
[0078] (6.2) Calculate the frame signal b s The short-time spectral entropy H of (m) s :
[0079]
[0080] where Y s (k) is the framed signal b s (m) the spectrum of the frequency component of the kth spectral line after fast Fourier transform;
[0081] (6.3) According to the short-time logarithmic energy LE s and short-time spectral entropy H s Calculate the frame signal b s (m) short-time energy entropy ratio Ef i (s):
[0082]
[0083] (6.4) Repeat steps (6.1) to (6.3) until the short-time energy entropy ratio Ef of all final frame signals is i (s) Calculation completed;
[0084] (6.5) For all framed signals b s (m) short-time energy entropy ratio Ef i (s) Take the average value and get the average value of its short-time energy entropy ratio Efi:
[0085] Efi=mean(Ef i (s)).
[0086] Step 7: Perform adaptive threshold endpoint detection on the segmented speech signal.
[0087] Reference Figure 2 , the specific implementation of this step is as follows:
[0088] (7.1) The short-time energy entropy ratio Ef of each frame signal i (s) is compared with the preset retention value EfB to obtain the preliminary presence or absence of the speech frame of the signal segment:
[0089] If the short-time energy entropy ratio of the signal of more than 5 consecutive frames is Ef i If (s) is greater than or equal to EfB*(1+δ), it is regarded as a preliminary continuous speech frame, and the average short-time energy entropy ratio EfA of the preliminary continuous speech frame is calculated, and step (7.2) is executed, where δ is the preliminary judgment parameter of the speechless segment obtained by the experiment.
[0090] Otherwise, it is considered that all the frames in the signal segment are silent frames;
[0091] (7.2) Set the silent frame parameter: 0≤λ1≤1, and calculate the final endpoint detection threshold EfI:
[0092] EfI=Efi*λ1+(EfB+EfA)*(1-λ1) / 2
[0093] (7.3) The short-term energy entropy ratio Ef of the initial continuous speech frames i (s) is compared with the threshold to obtain the final detection result:
[0094] If the short-time energy entropy ratio of the signal of more than 4 consecutive frames is Efi (s) is greater than or equal to EfI, then the detected preliminary continuous talking frame is regarded as the final continuous talking frame, and EfB=EfB*λ2+(1-λ2)*EfI, where 0≤λ2≤1;
[0095] Otherwise, the initial continuous speech frames are all regarded as no speech frames, and EfB=EfBλ3+(1-λ3)*EfI, where 0≤λ3≤1, and λ2≤λ3;
[0096] (7.4) Repeat steps (7.1) to (7.3) until the detection of this segmented speech signal is completed.
[0097] Step 8: Estimate the segment received signal x i (n) signal-to-interference-noise ratio.
[0098] (8.1) Calculate the segmented received signal x i (n) signal power spectrum PF i (f r ):
[0099]
[0100] Among them, f r =(-L1 / 2:L1 / 2-1)*(Fs / L1) is the signal power spectrum PF i (f r ) all frequencies;
[0101] (8.2) According to the signal power spectrum PF i (f r ) Estimate the signal carrier power P zbi :
[0102] P zbi =PF i (f zb -1)+PF i (f zb )+PF i (f zb +1)
[0103] where f zb is the carrier frequency;
[0104] (8.3) According to the endpoint detection result obtained in step 6, the proportion of speech frames in the current segmented speech signal to the total number of frames α is calculated:
[0105]
[0106] Among them, strb d is the d-th segment speech signal x i(n) The last frame of the last continuous conversational frame detected in d is the start frame corresponding to the end frame, and D is the number of the final continuous frames with conversations detected;
[0107] (8.4) Estimate the speech signal power in the received signal power spectrum according to the ratio α of speech frames in the segmented speech signal to the total number of frames: in, The power ratio of the carrier signal to the speech signal in a frame of received signal with speech but no interference obtained in the experiment;
[0108] (8.5) By P hy With P zbi Estimated segment received signal interference and noise ratio: ESIR i =10*log((P zbi +P hy ) / (P i -P zb )), where P i The received signal x for the i-th segment i (n) total power.
[0109] Step 9: Estimate the signal-to-interference-noise ratio of the received signal.
[0110] (9.1) Repeat steps 5 to 8 until all segments receive signal x i The signal to noise ratio (n) is estimated, and the signal to noise ratio estimation results of all segment received signals are ESIR1, ESIR2…ESIR i …、ESIR N1 ;
[0111] (9.2) Take the maximum estimation result ESIR of the signal-to-interference-plus-noise ratio estimation results of all segmented received signals and use it as the signal-to-interference-plus-noise ratio estimation result of the received signal: ESIR = max(ESIR1, ESIR2…ESIR i …、ESIR N 1).
[0112] Step 10: Perform quality assessment on the received signal based on the signal to interference noise ratio estimation result.
[0113] 10.1) The quality assessment criteria are as follows:
[0114] The interference-free quality assessment standard is above 6dB;
[0115] The weak interference quality assessment standard is -5dB to 6dB;
[0116] The medium interference quality assessment standard is -9dB to -5dB;
[0117] The strong interference quality assessment criterion is below -9 dB;
[0118] 10.2) Compare the estimated result of the received signal SINR, ESIR, with the assessment criterion to obtain the received signal quality assessment result.
[0119] If the estimated result of the SINR, ESIR > 6 dB, it is regarded as no interference;
[0120] If the estimated result of the SINR, -5 dB < ESIR < 6 dB, it is regarded as weak interference;
[0121] If the estimated result of the SINR, -9 dB < ESIR ≤ -5 dB, it is regarded as medium interference;
[0122] If the estimated result of the SINR, ESIR < -9 dB, it is regarded as strong interference.
[0123] The effects of the present invention can be further illustrated by the following simulation results:
[0124] I. Simulation parameters
[0125] Set the interference type, interference bandwidth, input SINR range, communication signal type, and number of experiments, as shown in Table 1
[0126] Table 1 Simulation parameters
[0127] Interference Type NoiFM, QPSK, 2FSK, BPSK Interference bandwidth 5k, 10k, 25k, 50k, 100k Input signal-to-interference-noise ratio range --10dB~10dB Communication signal type AM Number of experiments 50 times
[0128] II. Simulation content
[0129] Simulation 1, use the present invention to estimate the SINR of the mixed signals of 5 different bandwidths of the NoiFM interference type respectively, and the results are as Figure 3
[0130] From Figure 3 It can be seen that for the mixed signals of 5 different bandwidths of the NoiFM interference type within -10 dB to 10 dB, the SINR estimation result graphs almost overlap, indicating that the present invention can perform relatively accurate SINR estimation on the mixed signals of multiple different bandwidths of the NoiFM interference type within -1 dB to 10 dB.
[0131] Simulation 2, use the present invention to estimate the SINR of the mixed signals of 5 different bandwidths of the QPSK interference type respectively, and the results are as Figure 4
[0132] From Figure 4It can be seen that for mixed signals of QPSK interference types with 5 different bandwidths within -10dB~10dB, their signal to interference noise ratio estimation result graphs are almost overlapped, indicating that the present invention can perform relatively accurate signal to interference noise ratio estimation for mixed signals of QPSK interference types with multiple different bandwidths within the signal to interference noise ratio of -1vdB~10dB.
[0133] Simulation 3, using the present invention to estimate the signal to noise ratio of mixed signals of 2FSK interference types with five different bandwidths, the results are as follows: Figure 5
[0134] from Figure 5 It can be seen that for mixed signals of 2FSK interference types with 5 different bandwidths within -10dB~10dB, their signal to interference noise ratio estimation result graphs are almost overlapped, indicating that the present invention can perform relatively accurate signal to interference noise ratio estimation for mixed signals of 2FSK interference types with a plurality of different bandwidths within the signal to interference noise ratio range of -10dB~10dB.
[0135] Simulation 4, using the present invention to estimate the signal to noise ratio of mixed signals of BPSK interference types with five different bandwidths, the results are as follows: Figure 6
[0136] from Figure 6 It can be seen that for mixed signals of BPSK interference types with 5 different bandwidths within -10dB~10dB, their signal to interference noise ratio estimation result graphs are almost overlapped, indicating that the present invention can perform relatively accurate signal to interference noise ratio estimation for mixed signals of BPSK interference types with multiple different bandwidths within the signal to interference noise ratio range of -10dB~10dB.
[0137] The above simulation results show that the signal to noise ratio estimation results for different interference bandwidths of the present invention are similar, and are less affected by the interference bandwidth. Moreover, the error of the signal to noise ratio estimation result of the present invention relative to the set real mixed signal signal to noise ratio is no more than 1 dB, thereby providing an accurate signal to noise ratio estimation result for subsequent signal quality evaluation.
Claims
1. A radio station AM voice signal quality assessment method based on adaptive threshold, include: (1) Segment the received signal X(n) to obtain the segmented received signal x i (n)n=1,2···,L1,i=1,2···,N1,N1 is the number of segments, L1 is the signal length after segmentation; (2) Demodulate the received signal X(n) to obtain a speech signal A(n); (3) Segment the demodulated speech signal A(n) to obtain segmented speech signal a i (n); (4) For the segmented speech signal a i (n) Frame division and calculate the s-th frame signal b s (m) short-time energy entropy ratio Ef i (s), and average the short-time energy entropy ratio of all frame signals of this segment to obtain the segmented speech signal a i (n) average short-time energy entropy ratio Efi; (5) Preset the retention value EfB, and compare the short-time energy entropy ratio of each frame signal to Ef i (s) is compared with it to obtain the preliminary presence or absence of the speech frame of the signal: If the short-time energy entropy ratio of the signal of more than 5 consecutive frames is Ef i If (s) is greater than or equal to EfB*(1+δ), it is regarded as a preliminary continuous speech frame and step (6) is executed, wherein δ is a parameter for judging the speechless segment obtained by experiments. Otherwise, it is considered that all the frames in the signal segment are silent frames; (6) Calculate the average short-time energy entropy ratio EfA of the initial continuous frames with conversation and the final endpoint detection threshold EfI, and convert the short-time energy entropy ratio EfA of the frames with conversation into i (s) is compared with the threshold to obtain the final detection result: If the short-time energy entropy ratio of the signal of more than 4 consecutive frames is Ef i (s) is greater than or equal to EfI, then the detected continuous talking frame is regarded as the final continuous talking frame, and EfB=EfB*λ 2 +(1-λ 2 )*EfI, where 0≤λ 2 ≤1. Otherwise, the initial continuous frames with speech are all regarded as frames without speech, and EfB = EfBλ 3 +(1-λ 3 )*EfI is updated, where 0≤λ 3 ≤1, and λ 2 ≤λ 3 ; (7) Repeat steps (5) to (6) until the detection of the speech signal of the current segment is completed, and calculate the ratio α of the speech frames with speech in the current segment of the speech signal to the total number of frames; (8) Estimate the signal-to-interference-plus-noise ratio (ESIR) of this segment using the ratio α i : (8a) Calculate the segmented received signal x i Power spectrum PF of (n) i (f r ), through the power spectrum PF i (f r ) Estimate the segment received signal x i (n) Carrier power: P zbi =PF i (f zb -1)+PF i (f zb )+PF i (f zb +1), where f zb is the corresponding position of the carrier frequency; (8b) The carrier power estimation result P is calculated using the ratio α. zbi Correction is performed to obtain the corrected speech signal power in, The power ratio of the carrier signal to the speech signal in a frame of received signal with speech but no interference obtained in the experiment; (8c) By P hy With P zbi Estimated segment received signal interference and noise ratio: ESIR i =10*log((P zbi +P hy ) / (P i -P zb )), where P i The segmented received signal x i (n) total power; (9) Repeat steps (4) to (8) until all segments receive signal x. i The signal to noise ratio (n) is estimated, and the signal to noise ratio estimation result of all segment received signals is ESIR 1 ESIR 2 ...ESIR i ...、ESIR N1 ; (10) Select the maximum value ESIR among all the segment signal estimation results = max(ESIR 1 ESIR 2 ...ESIR i ...、ESIR N1 ) as the signal-to-interference-noise ratio estimation result of the received signal, and the signal quality of the received signal is judged according to the result: If the estimated result ESIR>6dB, it is considered that there is no interference; If -5dB<ESIR≤6dB, it is considered as weak interference; If -9dB<ESIR≤-5dB, it is considered as medium interference; If ESIR≤-9dB, it is considered as strong interference.
2. The method according to claim 1, Features: Step (1) segments the received signal X(n) by segmenting the received signal X(n) in the time domain according to the set signal length L1, into N1 segmented signals. The i-th segmented received signal is expressed as follows: x i (n)=X((i-1)*L1:i*L1)。 3. The method according to claim 1, Features: In step (2), the received signal X(n) is non-coherently demodulated as follows: (2a) The received signal X(n) is combined with the carrier signal X zb (n) = cos(2π*fc*n) to obtain the multiplication signal, A”(n): A”(n)=X(n)*X zb (n) Where fc is the carrier frequency of the received signal X(n); (2b) performing a low-pass filter on the multiplication signal A"(n) with a frequency of fd*2 / Fs to obtain a filtered signal A'(n), where fd is the bandwidth of the carrier signal; (2c) Remove the DC component from the filtered signal A'(n) to obtain the demodulated speech signal A(n): A(n)=A(n)'-mean(A(n)').
4. The method according to claim 1, Features: Step (3) segments the demodulated speech signal A(n) according to the set signal length L1 in the time domain of the received signal A(n), and divides the received signal into N1 segments of segmented signals. The segmented speech signal of the i-th segment is expressed as follows: to i (n)=A((i-1)*L1:i*L1)。 5. The method according to claim 1, Features: Step (4) performs segmented speech signal a i (n) Framing is performed as follows: (4a) In the segmented speech signal a i (n) Overlapping frames are performed on the time domain length to obtain the preliminary frame signal bl s (m) is represented as follows: bl s (m)=a i ((s-1)*inc:(s-1)*inc+wlen) Wherein, frame length wlen = L1 / 50, frame shift inc = wlen / 2, m = 1, 2...wlen, s = 1, 2...fn, fn is the number of sub-frames; (4b) Window all preliminary framed signals to obtain the final framed signal b s (m): b s (m)=bl s (m)*w(m) in is the Hamming window.
6. The method according to claim 1, Features: Step (4) Calculate the S-th frame signal b s (m) short-time energy entropy ratio Ef i (s), implemented as follows: (4c) Calculate the frame signal b s The short-time logarithmic energy LE of (m) s : Where a is a preset constant; (4d) Calculate the frame signal b s The short-time spectral entropy H of (m) j : where Y s (k) is the framed signal b s (m) the spectrum of the frequency component of the kth spectral line after FFT transformation; (4e) According to the short-time logarithmic energy LE s and short-time spectral entropy H s Calculate the frame signal b s (m) short-time energy entropy ratio Ef i (s):
7. The method according to claim 1, Features: Step (5) presets the retention value EfB, which is implemented as follows: (5a) Mix Ns frames of different silent frame signals with the interference signal of the current test environment of the same length to obtain Ns frames of mixed signals: xh jl (n) jl=1,2,...,Ns; (5b) Calculate the mixed signal xh jl The average short-time energy entropy ratio of (n) obtains the retention value EfB.
8. The method according to claim 1, Features: In step (6), the final endpoint detection threshold EfI is calculated as follows: EfI=Efi*λ 1 +(EfB+EfA)*(1-λ 1 ) / 2,0≤λ 1 ≤1 Among them, Efi is the segmented speech signal a i (n), EfA is the average short-time energy entropy ratio of the current preliminary continuous talk frames, and EfB is the current preset reserved value.
9. The method according to claim 1, Features: In step (7), the segmented speech signal x is calculated i (n) The ratio of the number of frames with conversations to the total number of frames is α, and the formula is as follows: where strb d is the speech signal x of this segment i (n) The last frame of the last continuous conversational frame detected in d is the starting frame corresponding to the last frame, and D is the number of the final continuous frames with conversations detected.
10. The method according to claim 1, Features: In step (8a), the segmented received signal x is calculated i Power spectrum PF of (n) i (f r ), the formula is as follows: Among them, f r =(-L1 / 2:L1 / 2-1)*(Fs / L1) is the signal power spectrum frequency, and Fs is the sampling rate of the received signal X(n).
Citation Information
Patent Citations
Output-based objective voice quality evaluation method
CN107293306A
Signal quality evaluation device and signal quality evaluation method
CN114124257A
