161 results about "Speech denoising" patented technology

The invention discloses a training method and apparatus of a neural network model, and a speech denoising method and device. The training apparatus trains a neural network by performing a neural network training method to obtain a noise-separated neural network model; the speech denoising device performs a speech denoising method to obtain scene noise data or speech data in a tested speech sampleby using the noise-separated neural network model; and then the scene noise data or speech data in the tested speech sample are removed. Therefore, a technical problem that the existing denoising method needs to be manually adjusted according to the specific condition of the input data when different to-be-denoised data are inputted and thus the universality of the denoising method is poor in theprior art is solved.

The invention discloses a sound source positioning method based on a dual-microphone array. The method comprises the following steps that: S1: firstly, converting a simulated sound signal into two paths of digital sound signals; S2: enabling the digital sound signals to pass through a sound detection module, and sending a detected part with human voice speaking into a denoising processing module;S3: after a human voice signal is subjected to the denoising processing, calculating the mutual correlation information of two paths of signals to obtain the time delay difference of two paths of signals; and S4: finally, utilizing the size of the microphone array and the calculated time delay difference to calculate the position information of the sound source. A calculation amount to which the method relates is small, implementation cost is low, voice denoising processing and score time delay estimation processing are added, and therefore, the robustness of sound source positioning and the accuracy of a positioning angle are improved.

The embodiment of the invention discloses a voice processing method, a voice processing device and a device for processing voice. The embodiment of the method comprises the following steps of performing time-frequency analysis on noisy voice to obtain the frequency spectrum of the noisy voice in complex domains; inputting the frequency spectrum of the noisy voice in the complex fields into a pretrained time-frequency shielding prediction model; obtaining a prediction value of time-frequency masking of the noisy voice in the complex domains; multiplying the prediction value by the frequency spectrum of the noisy voice in the complex domains; generating the frequency spectrum of target voice in the noisy voice in the complex domains; and synthesizing the target voice on the basis of the frequency spectrum of the target voice in the complex domains. The embodiment has the advantages that the voice distortion degree is reduced, and the voice noise reduction effect is improved.

The invention relates to a voice noise reduction method and device based on signal autocorrelation. The method comprises steps that 1), a noise-containing simulation voice signal is acquired through avoice reception module; 2), the simulation signal is converted by an A / D conversion module into a digital signal; 3), the digital signal is enframed through a voice enframing module; 4), autocorrelation operation of the signals after enframing is carried out through an autocorrelation de-noising module; 5), FFT analysis on an autocorrelation function of each noise-containing signal in a time-delay domain is carried out through an FFT analysis module; 6), threshold de-noising processing for FFT transformation is carried out through a threshold noise reduction module; 7), re-sampling of a frequency domain autocorrelation signal is carried out through a re-sampling and phase matching module, and phase matching of the frequency domain autocorrelation signal and the noise-containing signals iscarried out; 8), inverse FFT transformation is carried out by an inverse FFT transformation module through utilizing a frequency domain amplitude after processing and the phase, and the signals are returned to the time domain; and 9), recombination of each signal after de-noising is carried out through a frame recombination module to acquire voice signals after de-noising. The method is advantaged in that language sharpness can be substantially improved under the -20dB signal to noise ratio strong noise environment.

The invention discloses a speech denoising method based on a time frequency domain. An input signal is divided into data blocks according to a sampling sequence, and each data block is converted into a frequency domain signal through discrete Fourier transform. A time frequency domain is formed by the data which is subjectd to discrete Fourier transform. A time frequency domain signal is subjected to statistical analysis and identification, a speech signal and a noise signal can be roughly separated, and the modeling of the two signals is carried out and characteristic values are extracted. According to the extracted characteristic values, a filter is designed and adjusted in real time, and the time frequency domain signal is filtered. Finally, the filtered time frequency domain signal is subjected to discrete Fourier transform to be a time domain audio signal. The method has the advantages that the system mechanical structure design and production process are simplified, the system cost is reduced, the frequency domain filter can be designed and adjusted in real time, and the input voice signal is filtered to achieve purpose of denoising, the cumbersome filter operation is simplified as single frequency domain gain multiplication operation, the processing calculation amount is reduced, and the processing efficiency is improved.

The invention relates to the technical field of communication noise reduction, in particular to a single microphone voice noise reduction method. According to the method, time slots with no user voice input during a conversation process are identified, and voice signals collected by a microphone during the time slots with no user voice input are all background noise signals, so that the time slots are used for obtaining the background noise signals, then noise filtering processing is conducted on the voice signals during the conversation process according to the background noise signals, and therefore the noise disturbance problem can be solved by only one microphone, and the cost and design difficulty of mobile communication equipment are reduced.

The invention discloses a speech denoising device and a method based on wavelet transform; the device comprises a speech acquiring device, a pre-processing device, a wavelet transform device and a signal output device. The working method mainly comprises the following steps: step 1, performing the wavelet transform on wavelets with noise to obtain wavelet coefficients on various scales; step 2, performing soft or hard threshold function processing on the wavelet coefficients in various scales to obtain estimated values of wavelet coefficients of original signals; and step 3, obtaining denoised signals by the processed wavelet coefficients through wavelet inverse transform. The device and the method have the benefits that the wavelet analysis can analyze outlines of the signals, and can analyze details of the signals, so that the noise is effectively restrained; the loss of speech section information is reduced, and important significance is brought to future research on speech denoising algorithm; and the speech noise can be effectively removed so as to achieve the purpose of recovering the original speech signals.

The invention provides a vehicle-mounted voice intelligent Bluetooth integration device and method. A first record is obtained through microphone or microphone array recording and LineIn recording, and a second record is obtained through echo cancellation, voice enhancement and voice noise reduction; and the second record is subjected to local voice recognition through an offline recognition unitor cloud recognition through a mobile phone voice assistant unit, an interaction message or an equipment control message is obtained after recognition, the interaction message is executed by the mobile phone voice assistant unit to return audio output, and the equipment control message is transmitted to an automobile control center to be processed. Through intelligent voice recognition processing,the common vehicle-mounted Bluetooth intellectualization is promoted; the vehicle-mounted Bluetooth is connected with the mobile phone, the mobile phone serves as the center, the vehicle-mounted Bluetooth can easily have the AI intelligent interaction capacity, voice interaction between a user and the vehicle-mounted equipment does not need to be conducted manually, the vehicle-mounted Bluetoothequipment can obtain the capacity and resources recognized by the third-party cloud, then the device becomes truly intelligent, and the vehicle-mounted Bluetooth is endowed with new competitiveness.

The invention discloses a speech denoising method and a device thereof. The method comprises the following steps: receiving a set of noisy speech signals; extracting BFCC features of each noisy speechsignal in the set; inputting the BFCC features into a neural network GRU for training to generate a recurrent neural network RNN model; receiving a speech signal to be denoised; extracting BFCC features of the speech signal to be denoised; inputting the BFCC features of the speech signal to be denoised into the RNN model; determining a noise spectrum in the speech signal to be denoised based on the RNN model; determining a gain compensation parameter of the speech signal in the speech signal to be denoised according to the noise spectrum in the speech signal to be denoised; and generating a denoised speech signal based on the gain compensation parameter of the speech signal and a fundamental tone signal in the speech signal to be denoised. The technical problem that in the prior art speech signals are difficult to extract, and the quality of extracted speech is poor are solved.

The embodiment of the invention provides an audio noise reduction method and device, electronic equipment and a medium. The audio noise reduction method provided by the invention comprises the following steps: separating a first accompaniment signal and a first voice signal from an audio signal to be subjected to noise reduction; filtering the first accompaniment signal to obtain a second accompaniment signal; filtering the first voice signal to obtain a second voice signal; and synthesizing a noise reduction audio signal according to the second accompaniment signal and the second voice signal. The invention provides an audio noise reduction method. Noise reduction is carried out on an accompaniment signal and a voice signal after separation; and finally, the noise-reduced accompaniment signal and the voice part signal are synthesized again. The method is advantaged in that noise is effectively removed, better voice quality is realized, severe distortion of the processed audio signal caused by inhibition of some high frequency portions of the music accompaniment by the voice noise reduction algorithm can be effectively avoided, and better voice experience is provided for game or live broadcast users.

The invention discloses a hierarchical speech noise reduction recognition method and system under a noise environment. The method comprises the following steps that a common text for communication bytelephone is set, and text entry is carried out; the set text is used for establishing a text database; according to the text, recording of pure speech and all kinds of noise is carried out; linear combination is carried out on the pure speech and all kinds of the noise to establish a speech database; endpoint detection is carried out on noisy speech to detect a front endpoint and rear endpoint ofeffective speech, and the effective speech is extracted; the effective speech is trained through a hierarchical method to obtain a speech enhancement deep neural network; the effective speech is subjected to speech enhancement, and then acoustic features of the effective speech are extracted; the acoustic features are used for training an acoustic model; the text is used for training a language model; the acoustic model and the language model are used for searching to obtain an optimum recognition result sequence. The hierarchical speech noise reduction recognition method and system under thenoise environment can improve the accuracy degree of recognition of original speech of a customer through noise reduction processing of the original speech of the customer.

The embodiment of the invention discloses a noise reduction method and device. The method comprises the following steps: training a time domain characteristic value and a target value of the trainingdata through a deep learning training model constructed based on a frequency domain; the server can send the obtained model parameters to the terminal device, so that the terminal device updates the parameters of the first voice noise reduction model after receiving the model parameters, and uses the updated first voice noise reduction model to perform noise reduction processing on voice information input by a user. According to the embodiment of the invention, parameters of a first voice noise reduction model in terminal device are updated by using model parameters obtained by training of a deep learning training model, the terminal device can use the model parameters obtained by training of the deep learning training model to perform noise reduction processing on the voice information input by the user, so that the voice information obtained by noise reduction processing is more accurate, and the user experience is improved.

The invention discloses a voice noise reduction method and relates to the field of machine learning, and the method is used for solving the problems of large calculation amount and large resource occupation of the existing voice noise reduction. The method comprises the following steps of obtaining voice data; preprocessing the voice data, and extracting multi-dimensional features of the preprocessed voice data; inputting the multi-dimensional features into a preset voice noise reduction model to obtain a frequency band gain coefficient; dividing the voice data into a plurality of frequency bands, and filtering noise data in the frequency bands according to the frequency band gain coefficient; and recovering the filtered voice data into a voice data stream and outputting the voice data stream. The invention further discloses a voice noise reduction device, electronic equipment and a computer storage medium. According to the invention, the frequency band gain coefficient is calculated to realize voice noise reduction.

The method is suitable for the technical field of voice noise reduction, and provides a voice noise reduction method, a voice noise reduction device and terminal equipment. The method comprises the following steps: obtaining first voice data and second voice data, fusing the first voice data and the second voice data, obtaining training data, training a generative adversarial network according tothe first voice data and the training data, obtaining a trained generative adversarial network model, performing noise reduction on voice recognition data to be processed through the trained generative adversarial network model, and obtaining target voice recognition data. According to the invention, the noise interference in the voice can be reduced, the voice recognition accuracy is improved, and the voice data can accurately transmit the information of the user.

The embodiment of the invention provides a voice reduction method and device, a mobile terminal, and a computer readable storage medium. The method is used for the mobile terminal, and the mobile terminal is connected with an earphone. The method comprises the steps: collecting a first audio signal; receiving a second audio signal transmitted by the earphone; detecting whether the mobile terminal is shaded or not; extracting a target noise signal from the first audio signal and / or the second audio signal based on a detection result; carrying out the noise reduction of a third audio signal obtained by the mobile terminal through the target noise signal, and outputting a fourth audio signal. The third audio signal is an audio signal collected by the earphone or an audio signal to be transmitted to the earphone. According to the embodiment of the invention, the method achieves a purpose of dynamically selecting a proper noise reduction scheme according to the environment, improves the noise pickup precision, and improves the noise reduction effect.

The invention relates to a real-time voice denoising method under a short-distance conversation scene used for a double-microphone mobile phone. The method comprises the following steps: receiving noisy voice signals of a main channel and an auxiliary channel, wherein the noisy voice signals comprise voice and background interfering noises; extracting original acoustic features from the noisy voice signals; inputting the original acoustic features into a convolutional recursive network (CRN) for simulating temporal dynamics of the voice, and carrying out masking estimation, thus obtaining a phase-sensitive mask PSM; and applying the estimated phase-sensitive mask PSM to the noisy voice signals of the main channel, so that wave signals of the denoising voice are re-synthesized. According tothe technical scheme, the performance is excellent, conversation noises can be effectively reduced under the short-distance conversation scene, and compared with the prior art, the technical scheme has the advantages of having stronger practicability, and having no dependence on the noises and a speaker.

The invention belongs to the field of voice noise reduction, and discloses a single-track voice noise reduction method, system and device, and a readable storage medium, and the method comprises the steps: obtaining to-be-denoised single-track voice; constructing an initial noise reduction model based on an LSTM neural network; obtaining a preset number of enhanced training samples, and training the initial noise reduction model by adopting the preset number of enhanced training samples to obtain a noise reduction model; and denoising the to-be-denoised single-track voice through the noise reduction model, so that a human voice audio can be obtained. The noise reduction process is not limited by double channels, noise reduction processing of any single-track voice can be realized, model training is performed on the basis of the LSTM neural network on the basis of the characteristic that the noise types of the whole time sequence in the same voice are approximately the same, the noise law of the whole time sequence is convenient to learn, and therefore, a good noise reduction effect is achieved. Meanwhile, based on the complexity of the noise influence factors, the initial noise reduction model is trained again by the enhanced training samples, and the noise reduction effect of the noise reduction model is further improved.

The invention provides an audio signal processing method and device, equipment and a storage medium. The method comprises the following steps: acquiring an audio signal; determining a noise feature corresponding to the audio signal by using a pre-trained noise reduction model; processing the noise feature corresponding to the audio signal to obtain a voice feature corresponding to the audio signal; and processing the voice feature corresponding to the audio signal to obtain a denoised target audio signal corresponding to the audio signal. Audio signals are input into a pre-trained noise reduction model, the noise reduction model comprises a noise classifier and a voice noise reduction device, the noise classifier is used for determining noise features corresponding to the audio signals, and the voice noise reduction device is used for determining voice features corresponding to the audio signals according to the noise features corresponding to the audio signals. Therefore, the voice noise reducer can utilize the noise features to assist in voice noise reduction, the voice noise reducer is better guided to extract the voice features, and a better voice noise reduction effect is achieved.

The invention relates to the technical field of audio processing, and discloses a voice noise reduction method. The method comprises the following steps: acquiring voice data, inputting the voice data into a preset standard scene recognition model, and determining a voice scene corresponding to the voice data, the standard scene recognition model being obtained by training according to a noise sample set in each scene, and selecting a preset noise reduction model corresponding to the voice scene, and performing noise reduction on the voice data. In addition, the invention also discloses a voice noise reduction method and device, equipment and a storage medium. According to the invention, the accuracy of voice noise reduction can be improved.

The invention discloses a single-channel voice simultaneous noise reduction and reverberation removal system. The system comprises a voice noise reduction module, a voice reverberation removal moduleand a joint training module. The voice noise reduction module trains a deep embedded feature extractor by using a deep clustering algorithm, extracts deep embedded features from mixed voice signals, and maps the input mixed voice into a noise-free embedded space, so that the deep embedded features do not contain noise and have great discrimination for reverberation and direct sound; the voice reverberation removal module is connected with the voice noise reduction module and is used for removing reverberation voice signals from the deep embedded features and estimating direct sound of a cleantarget, so that the purposes of voice noise reduction and reverberation removal are achieved; and the joint training module is connected with the voice noise reduction module and the voice reverberation removal module respectively, and is used for jointly optimizing the voice noise reduction module and the voice reverberation removal module, and improving the quality and intelligibility of the enhanced voice.

The invention provides a voice noise reduction method based on spectral subtraction and wavelet transform. The method comprises the following steps of removing a direct current component of an input pure voice signal; performing normalization processing on an amplitude value; superimposing white gaussian noise on the processed voice signal so as to obtain the voice signal with noise; setting a leading no-word section; calculating the frame number of the leading no-word section; processing the voice signal with noise by spectral subtraction; performing wavelet transform on the obtained voice signal subjected to spectral subtraction to obtain a wavelet signal with a discrete telescopic factor and a discrete translation factor; according to a low-frequency component in the wavelet signal anda high-frequency component of each decomposition layer, obtaining a low-frequency coefficient and a high-frequency coefficient of each decomposition layer; respectively processing the high-frequency coefficient of each decomposition layer by a zero setting processing method and a soft threshold processing method so as to obtain the processed high-frequency coefficient of each decomposition layer;and performing inverse wavelet transform to obtain a final noise reduction voice signal. The music noise can be obviously reduced; in addition, only little high-frequency distortion exists; and the noise reduction capability is further improved.

The invention provides a voice noise reduction method and device based on artificial intelligence, electronic equipment and a computer readable storage medium. The method comprises: acquiring a voice signal; obtaining a constraint condition corresponding to the voice signal; and calling a noise reduction model matched with the constraint condition to carry out noise reduction processing on the voice signal to obtain a noise-reduced voice signal. According to the invention, the noise reduction effect and efficiency of noise reduction processing can be improved.

The invention relates to a voice denoising processing method, a voice denoising processing device, a computer device and a storage medium. The voice denoising processing method comprises the followingsteps: when detecting that the distance between a voice collecting device and a target object achieves a preset value, acquiring a noisy voice signal collected by the voice collecting device, and carrying out frequency division processing on the noisy voice signal, so that a low-frequency-band signal is obtained; acquiring a magnitude spectrum and a phase spectrum of the low-frequency-band signal; acquiring a modulation domain signal corresponding to the magnitude spectrum; processing a magnitude spectrum or a power spectrum of the modulation domain by adopting a spectrum subtraction method,thus obtaining the magnitude spectrum or power spectrum of the modulation domain after denoising; compensating the phase spectrum of the modulation domain, thus obtaining the phase spectrum of the modulation domain after compensating; and based on the phase spectrum of the modulation domain after compensating, the magnitude spectrum of the modulation domain after denoising, and the phase spectrumof the low-frequency-band signal, acquiring a low-frequency-band signal after denoising, wherein the low-frequency-band signal after denoising is used for synthesizing a denoising voice signal. Due toa smooth over-subtraction factor adopted by the spectrum subtraction method, the problem that abundant voice information is eliminated due to the oversize smooth over-subtraction factor is avoided.

The embodiment of the invention provides a method for training a grouping model for voice grouping, and a voice noise reduction method. The voice noise reduction method comprises the steps: B1, obtaining an input audio, and extracting the grouping features of the input audio; B2, inputting the grouping features of the input audio into a grouping model, and predictively outputting the label of theinput audio; and B3, according to the label of the input audio, taking the transcription text obtained by the audio processed by the optimal noise reduction strategy of the group corresponding to thelabel as the transcription text of the input audio. According to the technical scheme provided by the embodiment of the invention, the group to which the input audio belongs can be predicted accordingto the anti-disturbance of the input audio, and the noise reduction strategy suitable for the group is selected for noise reduction, so that the transcription error rate can be reduced, and the transcription quality of a clean sample cannot be influenced while the anti-transcription is recovered to the original transcription.

The invention provides a voice noise reduction method and device, equipment and a medium, and belongs to the technical field of voice. According to the invention, the gain modulation information of the to-be-denoised voice is obtained according to the probability distribution of the to-be-denoised voice and the modulation gain function corresponding to each noise type, and then the to-be-denoisedvoice is denoised according to the gain modulation information. Because noise time domain characteristics, frequency domain characteristics and the like of different noise types are different, different noise reduction processes are needed; compared with the prior art in which the same noise reduction algorithm is used to reduce the noise of the to-be-denoised voice, the situation that the noise types of the noise contained in the to be denoised voice are different is considered; different gain modulation information is acquired according to different noise types of noise contained in the to-be-denoised voice, so that the denoising processes of different to-be-denoised voices are distinguished, more targeted denoising is realized, the denoising effect is improved, and the wake-up rate andthe recognition rate of the denoised voice are improved.

The invention discloses a short-wave channel speech enhancement method, and belongs to the technical field of communication. According to the invention, a neural network is applied to a short-wave communication voice denoising module, and the same convolutional neural networks based on mapping are selected according to the performance requirements and characteristics of Anti-fading Net and Denoising Net, so that a better noise suppression effect can be achieved; a convolutional neural network denoising model Denoising Net is combined with an existing classic unsupervised speech enhancement algorithm OMLSA, the denoising advantages of the two algorithms are reserved in an equal gain combination mode, and therefore the de-noised speech quality of a Denoising Net or OMLSA algorithm which is independently used is improved; and thirdly, by stacking the two convolutional neural networks with simple structures, a better denoising effect than that of a common neural network with a complex structure is obtained.

The invention relates to a voice noise reduction method and device, equipment and a storage medium. The method comprises the steps of obtaining a to-be-denoised spectrogram corresponding to a to-be-denoised voice; inputting the to-be-denoised spectrogram into a preset voice noise reduction model to be subjected to noise reduction processing, and acquiring a noise-reduced spectrogram, wherein the voice noise reduction model sequentially comprises a coding module, a frequency domain noise reduction module, a time domain noise reduction module, a decoding module and a mask enhancement and suppression module; and performing voice signal reconstruction on the frequency spectrogram after noise reduction to obtain a target voice. According to the invention, feature extraction, frequency domain noise reduction, time domain noise reduction, decoding and mask enhancement and suppression are performed in sequence by adopting the voice noise reduction model, effective noise reduction is realized by performing frequency domain noise reduction and time domain noise reduction in sequence, the noise reduction effect is improved, the accuracy of the obtained noise-reduced voice is improved, and the user experience is improved; and frequency domain noise reduction and time domain noise reduction are processed separately, decoupling of the time domain and the frequency domain is achieved, and streaming voice noise reduction is facilitated.

The invention provides a microphone array speech denoising enhancement method with a low signal-to-noise ratio effect. The microphone array speech enhancement algorithm based on improved perfect binary time-frequency masking of human ears solves the problem of poor performance of a microphone array speech enhancement algorithm in the prior art under the condition of low signal-to-noise ratio, themasking effect of the human ears is skillfully added to the microphone array speech denoising enhancement algorithm. Firstly, a primary-order acquisition enhanced voice signal is acquired, then a pre-estimation noise signal is judged according to a noisy signal and the primary-order acquisition enhanced voice signal, a threshold value is set through a signal-to-noise ratio relationship between primary-order voice acquisition enhancement and judged pre-estimation noise, a perfect binary time-frequency masking algorithm is calculated, and the enhanced voice signal is synthesized. The speech signal enhanced by the algorithm is higher in intelligibility and better in sound quality, and the performance advantage is more obvious under the scene of larger noise intensity and low signal-to-noise ratio.

The invention discloses an intelligent speech denoising communication method and device. To-b-denoised noisy speech data are received and preprocessed, thus the noisy speech data in a predetermined format are obtained, the noisy speech data in the predetermined format are sent to an adaptive filter bank denoising algorithm, a deep neural network denoising algorithm and a wiener filtering denoisingalgorithm to be denoised, a plurality of channels of denoised speech data are subjected to denoising quality assessment, a denoising quality score list is generated, and the channel of desnoised speech data, with the highest denoising quality score, in the denoising quality score list are selected to generate a final denoising speech file. Aiming at the characteristic that different technologiesare suitable for solving specific noise speech, final results of all methods can be selected to be combined to obtain the best denoising effect, meanwhile good adaptability is also achieved, the problem that a simplex denoising method cannot be suitable for noise situation diversity is solved, and various defects generated by using a simplex denoising technology are avoided.