Conference room sound system based on single-channel target spectrum adaptive FIR (Finite Impulse Response) filter

Abstract

The invention provides a conference room sound system based on a single-channel target spectrum adaptive FIR filter, and relates to the technical field of audio signal processing. According to the conference room sound system based on the single-channel target spectrum adaptive FIR filter, only one microphone is used for collection, deployment is simplified, a multi-microphone array and single-channel multi-speaker adaptation are not needed, and a user can import any'good hearing 'target spectrum envelope curve; when the system operates, voice collected by a single-channel microphone is used as a driving signal, errors between an output frequency spectrum and a target curve are calculated in real time, FIR filter coefficients are updated online, spectral errors between output and a target are calculated in each frame, FIR frequency response is adaptively updated in a frequency domain through NLMS / LMS, meanwhile, only spectral envelope is corrected, an original phase / F0 is not affected, and the frequency response of the FIR is corrected. And the suppression, pause and personality of the speaker are kept, so that the output spectrum is continuously close to the target curve and the rhythm and fundamental frequency of the speaker are kept.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of audio signal processing, in particular to a conference room sound system based on a single-channel target spectrum adaptive FIR filter. BACKGROUND

[0002] The conference room sound system is a sound system specially used in conference places, aiming to provide clear, distortion-free, echo-free and background noise-free sound quality, and to ensure that the sound in the conference can be accurately and efficiently transmitted to each participant. Its main components include microphones, mixing consoles, power amplifiers, loudspeakers, audio processors, cables and peripheral equipment (such as feedback suppressors, equalizers, etc.).

[0003] The existing conference room sound system measures the loudspeaker room response when the system starts, designs a one-time FIR equalizer to make the output frequency response flat; but the filter coefficients are fixed during operation and do not change with the acoustic characteristics or speech content of the speaker, that is, the existing conference room sound system only equalizes once after startup and does not update during operation, resulting in large differences in sound color and inconsistent listening experience among different speakers. Different ways are used in the prior art to correct them, but each has its own defects:

[0004] 1. Static equalization: only corrects the inherent frequency response of the loudspeaker + room, but the spectral envelope of different people differs greatly, and it is impossible to ensure that all speakers sound "good" and "full";

[0005] 2. Multi-microphone array + beamforming: although it can improve the sound pickup quality, it increases the hardware cost and deployment complexity, which is not conducive to general conference scenarios.

[0006] 3. AGC / gain control: mainly adjusts the overall volume, and it is difficult to fine-tune the spectral shape.

[0007] Therefore, there is an urgent need for a FIR that can import any "good" target curve and update in real time during operation, so that no matter who is speaking, the output can approach the target curve and retain the original rhythm and fundamental frequency, achieving the effect of "one system, multiple voices all sounding good".

[0008] To this end, the present application has developed a new conference room sound system based on a single-channel target spectrum adaptive FIR filter. SUMMARY

[0009] (I) Technical problems solved

[0010] In view of the deficiencies of the prior art, the conference room sound system based on a single-channel target spectrum adaptive FIR filter is provided, which solves the problem that the prior art conference room sound system measures the loudspeaker room response at system startup, designs a one-time FIR equalizer to make the output frequency response flat, but the filter coefficients are fixed during operation and do not change with the acoustic characteristics of the speaker or the speech content, that is, the existing conference room sound system is only equalized once after startup and is not updated during operation, resulting in large differences in sound color and inconsistent listening experience of different speakers.

[0011] (II) Technical solutions

[0012] To achieve the above object, the following technical solutions are used: the conference room sound system based on a single-channel target spectrum adaptive FIR filter comprises:

[0013] A single-channel microphone module is used to collect the speech signal (analog voltage) of the speaker in the conference room, and the speech signal output is sent to an AD (analog-to-digital conversion) circuit after being amplified by a microphone preamplifier.

[0014] A front-end preprocessing module is used to pre-emphasize (high-pass filter), frame, and window process the digital speech signal after AD, and calculate the FFT (fast Fourier transform) spectrum of each frame.

[0015] A target spectrum envelope storage and import module is used to load the user-defined target spectrum envelope curve T(k) in the dB format through a network port, and save it in the DSP internal SRAM or external Flash.

[0016] A real-time output estimation module is used to perform frequency domain multiplication on the current frame input signal X t (k) and the H t-1 (k) FIR frequency response obtained by the last iteration to obtain the output spectrum.

[0017] An error calculation and adaptive update module is used to calculate the error and update the frequency response according to the frequency domain NLMS algorithm.

[0018] An FIR coefficient reconstruction and smoothing module is used to generate new FIR filter coefficients by IDFT (inverse discrete Fourier transform) and windowing the updated frequency response H upd (k), and perform exponential smoothing on the new and old FIR coefficients.

[0019] Real-time convolution output module, which is used for convolution operation on next input frame, outputs equalized time domain signal, and drives loudspeaker through DA (digital-analog conversion) and power amplifier.

[0020] Preferably, an algorithm for calculating output spectrum in real-time output estimation module is provided:

[0021] (1) Take current frame FFT input signal X t (k);

[0022] (2) Perform frequency domain multiplication with frequency response H t-1 (k) obtained by iteration of previous frame:

[0023] Y t (k)=H t-1 (k).X t (k)。

[0024] Preferably, an algorithm for calculating error and frequency response in error calculation and adaptive update module is provided:

[0025] e t (k)=T(k)-|Y t (k)|;

[0026]

[0027] Wherein, u is learning rate, and epsilon is anti-zero.

[0028] Preferably, an algorithm for calculating FIR filter coefficient and exponential smoothing in FIR coefficient reconstruction and smoothing module is provided:

[0029]

[0030] h t [n]=αh t-1 [n]+(1-α)h new [n], alpha is in [0.8, 0.95].

[0031] Wherein, w h [n] is Hamming window, and L h is filter length.

[0032] Preferably, the single-channel microphone module is installed at the center of conference table or front end of room, so as to ensure good pickup of sound source of speaker.

[0033] (Three) beneficial effects

[0034] The application provides a conference room sound system based on single-channel target spectrum adaptive FIR filter.

[0035] The conference room sound system based on the single-channel target spectrum adaptive FIR filter simplifies deployment by using only one microphone to collect, does not need a multi-microphone array, is adaptive to single-channel multi-speakers, and a user can import an arbitrary "good sound" target spectrum envelope curve; the system drives a signal in real time by using a single-channel microphone to collect a voice, calculates an error between an output spectrum and a target curve in real time, and updates FIR filter coefficients online; each frame calculates a spectrum error between an output and a target, updates an FIR frequency response in a frequency domain through NLMS / LMS, simultaneously, only corrects a spectrum envelope, does not affect an original phase / F0, maintains a speaker's intonation and personality, makes an output spectrum continuously close to a target curve, and retains a speaker's rhythm and fundamental frequency. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The module structure diagram of the present application. DETAILED DESCRIPTION

[0037] As Figure 1 shown, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0038] Embodiment 1

[0039] The conference room sound system based on the single-channel target spectrum adaptive FIR filter, the system hardware of the present embodiment mainly consists of the following parts, and is connected with each other through a high-speed DSP bus or DMA (Direct Memory Access), including:

[0040] (1) a single-channel microphone module, the single-channel microphone module is used to collect a voice signal (analog voltage) of a speaker in a conference room, and the voice signal is output to an AD (analog-to-digital conversion) circuit after being sent through a microphone preamplifier;

[0041] Specifically, a single-directional condenser microphone is selected as the electrical equipment, and the single-directional condenser microphone is installed at the center of a conference table or at the front end of a room, so as to ensure good sound pickup of a sound source of a speaker;

[0042] In addition, the single-channel microphone module further includes an analog-to-digital conversion and clock module and a DSP processor and storage, wherein:

[0043] The analog-to-digital conversion (ADC) module is absolutely core. The microphone outputs an analog audio signal, which needs to be converted into a digital signal by the ADC for subsequent processing. A 24-bit ADC (analog-to-digital converter) can be selected. The clock module provides an accurate sampling clock for the ADC, ensuring accurate and stable sampling frequency. A 48 kHz sampling rate clock can be selected. The ADC and clock are the core "collection" parts, and the purpose is to convert the analog signal into a digital signal. Based on the above devices, a PCM (pulse code modulation) digital audio stream is output.

[0044] The DSP processor (digital signal processor) is used for real-time processing of the collected digital audio signal, such as noise reduction, echo cancellation, gain control, encoding and decoding (such as compression), etc. ADI 21489 DSP or equivalent models can be selected. The memory is used for temporary storage of collected audio data, or storage of DSP algorithm programs, configuration parameters, etc. A 256k SDRAM can be used for intermediate caching, and a 4MB Flash can be used to store target curves and program codes.

[0045] (2) The front-end preprocessing module is used for pre-emphasis (high-pass filtering), framing, windowing processing of the AD digital voice signal, and calculating the FFT (fast Fourier transform) spectrum of each frame.

[0046] Its role is to eliminate low-frequency gain prominence, frame and window to reduce spectral leakage, and provide stable spectral observation for subsequent adaptive update. The selected device is a pre-emphasis filter

first-order IIR high-pass, coefficient 0.97

[0047] (3) The target spectrum envelope storage and import module is used to load user-defined target spectrum envelope curve T(k) in dB format through the network interface (or the user uploads the target spectrum envelope in dB format through PC software). The interface is Ethernet, and it is saved in the internal SRAM or external Flash of the DSP.

[0048] (4) The real-time output estimation module is used to perform frequency domain multiplication on the current frame input signal X t (k) and the H t-1 (k) FIR frequency response obtained by the last iteration to obtain the output spectrum.

[0049] Specifically, an algorithm for calculating the output spectrum in the real-time output estimation module is provided:

[0050] First, take the current frame FFT input signal Xt (k), and the frequency response H t-1 (k) is obtained by frequency domain multiplication t (k) = H t-1 (k).X t (k).

[0051] (5) an error calculation and adaptive update module, which is configured to calculate an error and update a frequency response according to a frequency domain NLMS algorithm;

[0052] Specifically, an algorithm for calculating an error and a frequency response in the error calculation and adaptive update module is provided.

[0053] First, an amplitude spectrum |Y t (k)| is obtained, and an error e t (k) = T(k) - |Y t (k)| is calculated, and a frequency response is updated according to a frequency domain NLMS (Normalized Least Mean Square) algorithm: and an updated frequency response H upd (k) is output.

[0054] wherein parameters are a learning rate u = 0.1 and ε = 10 -8 .

[0055] (6) an FIR coefficient reconstruction and smoothing module, which is configured to generate new FIR filter coefficients by performing IDFT (Inverse Discrete Fourier Transform) and windowing on the updated frequency response H upd (k), and perform exponential smoothing on the new and old FIR coefficients.

[0056] The FIR coefficient reconstruction can ensure linear phase, control time domain delay, and make the filter frequency response tend to a target curve, and the exponential smoothing of the coefficients can avoid audio distortion caused by sudden changes in the filter coefficients.

[0057] Meanwhile, an algorithm for calculating FIR filter coefficients and exponential smoothing in the FIR coefficient reconstruction and smoothing module is provided.

[0058]

[0059] w h [n] is a Hamming window, and L h = 96 taps

[0060] Exponential smoothing:

[0061] h t [n] = αh t-1 [n] + (1 - α)hd [n], alpha = 0.9.

[0062] (7) Real-time convolution output module for convolution operation on the next input frame, outputting the equalized time domain signal, through DA (digital-analog conversion) and power amplifier driving the loudspeaker;

[0063] Wherein, the convolution implementation is FFT overlap-save, ensuring that the total delay of the system is less than or equal to 20 ms, the IFFT is 1024 points, the DA (digital-analog conversion) is 24-bit DAC, and the equalized time domain signal is sent to the power amplifier driving the loudspeaker

[0064] The present application is based on the theory of digital signal processing (DSP) and the principle of adaptive filtering, and realizes the following technical effects:

[0065] Frequency domain adaptive filtering - using NLMS algorithm to update frequency response in frequency domain, which is lower in calculation amount than time domain LMS and is convenient for designing linear phase filter, so as to realize real-time online adaptation, continuously iterate FIR in operation, solve the problem that traditional one-time equalization cannot adapt to different speakers;

[0066] Envelope-detail separation: only correct the amplitude spectrum, do not change the phase, ensure that the prosody and fundamental frequency characteristics of the speaker are kept as they are, import any target curve, users do not need to re-measure, directly load the "good sound template", and the system can be real-time fitted;

[0067] Exponential smoothing: smoothing is used at both ends of the update frequency and time domain, taking into account stability and response speed;

[0068] Single-channel simplified deployment: only one microphone is needed, without complex array or beamforming, reducing hardware cost;

[0069] Prosody and fundamental frequency retention: unlike the stiff sound of full amplitude correction, it ensures natural intonation and personalized tone.

[0070] To sum up, the technical effects achieved in actual use are as follows: (1) the average deviation between the output spectrum and the target curve can be controlled within ±1 dB, and the maximum deviation between frequency points is not more than ±2 dB, which greatly reduces the inconsistency of listening, continuously fits the target frequency response, and greatly improves the spectral consistency; (2) the comparative experiment using the short-time objective speech index STOI (Short-Time Objective Intelligibility) shows that the STOI score of different speakers is improved by an average of 12%, which significantly improves the intelligibility of conference remote speech, and the speech intelligibility and clarity are improved, and the PESQ (Perceptual Evaluation of Speech Quality) score is improved by about 0.4 points compared with the static scheme, which enhances the sound quality performance during communication and recording; (3) the total processing delay of the system is ≤20 ms, which meets the real-time interaction requirements of conferences, live broadcasts and the like, and when the speaker's tone or speed changes, it can converge to the new target spectrum within 0.5 s, ensuring that the best effect is continuously presented during continuous speech, and the adaptability and real-time performance are combined, and the delay is controllable; (4) only a single-channel microphone and a common DSP platform are required, without the need for a multi-microphone array or additional positioning hardware, the hardware cost is reduced by about 50% compared with a beamforming system, and no manual re-measurement or manual adjustment is required during operation, and the system can automatically run by importing the target curve with one key, reducing the on-site debugging frequency and maintenance workload of audio engineers, and reducing the deployment and maintenance costs.

[0071] Embodiment 2:

[0072] To achieve the same inventive purpose, the following alternative embodiments can also be used:

[0073] Alternative one: IIR adaptive filtering

[0074] An IIR (infinite impulse response) filter and a frequency domain or time domain RLS (recursive least squares) algorithm are used for updating, so that fewer coefficients are required but the phase nonlinearity needs to be processed.

[0075] Alternative two: sub-band gain adjustment

[0076] The frequency domain is divided into a plurality of sub-bands, and only the overall gain of each sub-band is updated by NLMS, so that the frequency point dimension is reduced and the calculation amount is reduced.

[0077] Alternative three: envelope estimation optimization

[0078] Instead of full-spectrum FFT, a Mel filter bank is used to estimate a small amount of sub-band energy, and then linear interpolation is used to return to the frequency point to achieve an approximate target response.

[0079] In conclusion, in use, after the system is powered on, the DSP reads the initial target curve and the predicted loudspeaker-room response from the Flash, calculates the initial H 0 (k) with h 0 [n]. After receiving a new frame of speech, the pre-processing, FFT, estimation output, error calculation, NLMS update, IDFT+smoothing, IFFT+output are executed in sequence, and the processing delay is controlled to be less than or equal to 20 ms. When the voice tone of a speaker or the environment changes, the error-driven update can converge within about 0.5 s, so that the output spectrum is continuously close to the target. The phase does not participate in the update, and only the amplitude is corrected, so that the fundamental frequency (F0) and the time-domain waveform characteristics of the original speech are completely preserved.

[0080] The application has novelty, creativity and significant industrial application value, and is suitable for various scenes such as conferences, live broadcasts and voice interaction.

[0081] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A conference room sound system based on a single-channel target spectrum adaptive FIR filter, characterized in that: include: A single-channel microphone module is used to collect the voice signal (analog voltage) of the speaker in the conference room. The voice signal output is sent to the AD (analog-to-digital conversion) circuit after passing through the microphone preamplifier; The front-end pre-processing module is used to perform pre-emphasis (high-pass filtering), framing, and windowing on the digital voice signal after AD, and calculate the FFT (Fast Fourier Transform) spectrum of each frame; Target spectrum envelope storage and import module, which is used to load the user-defined target spectrum envelope curve T(k) in dB format through the network port and save it in the DSP internal SRAM or external Flash; A real-time output estimation module is used to estimate the current frame input signal X t (k) and H obtained in the previous iteration t-1 (k) Perform frequency domain multiplication on the FIR frequency response to obtain the output spectrum; An error calculation and adaptive update module, wherein the error calculation and adaptive update module is used to calculate the error and update the frequency response according to the frequency domain NLMS algorithm; The FIR coefficient reconstruction and smoothing module is used to reconstruct the updated frequency response H upd (k) Generate new FIR filter coefficients through IDFT (inverse discrete Fourier transform) and windowing, and perform exponential smoothing on the new and old FIR coefficients; A real-time convolution output module is used to perform a convolution operation on the next input frame, output a time-domain signal after equalization, and drive a speaker through a DA (digital-to-analog converter) and a power amplifier.

2. The conference room sound system based on a single-channel target spectrum adaptive FIR filter according to claim 1 is characterized in that: Provides an algorithm for calculating the output spectrum in a real-time output estimation module: (1) Take the current frame FFT input signal X t (k); (2) The frequency response H obtained by iterative calculation of the previous frame t-1 (k) Perform frequency domain multiplication: Y t (k)=H t-1 (k).X t (k)。 3. The conference room sound system based on a single-channel target spectrum adaptive FIR filter according to claim 1 is characterized in that: Provides an algorithm for calculating the error and frequency response in the error calculation and adaptive update module: e t (k)=T(k)-|Y t (k)|; Where u is the learning rate and ε is the zero-proofing.

4. The conference room sound system based on a single-channel target spectrum adaptive FIR filter according to claim 1, characterized in that: Provides an algorithm for calculating FIR filter coefficients and exponential smoothing in the FIR coefficient reconstruction and smoothing module: h t [n]=αh t-1 [n]+(1-a)h new [n],α∈[0.8,0.95]; where w h [n] is the Hamming window, L h is the filter length.

5. The conference room sound system based on a single-channel target spectrum adaptive FIR filter according to claim 1, characterized in that: The single-channel microphone module is installed in the center of the conference table or at the front of the room to ensure good pickup of the speaker's voice source.

Images