Abstract: One disclosed example method includes a device receiving an audio signal recorded in a physical environment and applying a de-noise and de-reverberation model onto the audio signal to generate a cleaned audio signal. The de-noise and de-reverberation model is configured to remove noise and reverberation from the audio signal and is trained via a training process. The training process includes training the de-noise and de-reverberation model based on a trained de-noise teacher model and a trained de-reverberation teacher model. The training includes adjusting a portion of parameters of the de-noise and de-reverberation model based on values generated by the de-noise teacher model and the de-reverberation teacher model and then adjusting the parameters of the de-noise and de-reverberation model independently of the de-noise teacher model and the de-reverberation teacher model.

Abstract: A method performed by a processor of a computer system including a headset that is to be worn on a head of a user. The method drives a speaker of the headset with an input audio signal to output sound into an environment. The method determines that the speaker is at least partially covered with a cupped hand. In response to determining that the speaker is at least partially covered with the cupped hand, applying a gain to the input audio signal to reduce an output sound level of the speaker.

Abstract: A processing device according to an embodiment includes: a frequency characteristics acquisition unit configured to acquire frequency characteristics of at least one sound pickup signal; a smoothing processing unit configured to perform smoothing processing so as to generate second spectral data smoother than first spectral data based on the frequency characteristics; a first compression unit configured to calculate a first difference value corresponding to a difference between the second spectral data and the first spectral data in a first band, and to compress the second spectral data based on the first difference value; and a filter generation unit configured to generate a filter, based on the second spectral data.

Abstract: A method, computer program product, and computing system for receiving feature-based voice data associated with a first acoustic domain. One or more reverberation-based augmentations may be performed on at least a portion of the feature-based voice data, thus defining reverberation-augmented feature-based voice data.

Abstract: A method for controlling headphones. The headphones include a feedback microphone and a speaker. The method includes: obtaining an ear canal audio signal by collecting an audio signal in an ear canal by the feedback microphone; obtaining an audio signal feature parameter by conducting feature extraction on the ear canal audio signal; determining a tightness level of the headphones in a current wearing status; generating an audio feature to be identified according to the audio signal feature parameter and the tightness level; inputting the audio feature to be identified into a preset interactive operation identification model, and outputting an identification result, where the identification result includes an interactive operation identifier; determining a control instruction corresponding to the interactive operation identifier; and controlling a playback status of the speaker according to the control instruction.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a generative neural network to convert conditioning text inputs to audio outputs using energy scores.

Abstract: The present disclosure provides a microphone apparatus. The microphone apparatus may include a microphone and a vibration sensor. The microphone may be configured to receive a first signal including a voice signal and a first vibration signal. The vibration sensor may be configured to receive a second vibration signal. And the microphone and the vibration sensor are configured such that the first vibration signal may be offset with the second vibration signal.

Abstract: Methods, systems, and computer program products of automatic de-essing are disclosed. An automatic de-esser can be used without manually setting parameters and can perform reliable sibilance detection and reduction regardless of absolute signal level, singer gender and other extraneous factors. An audio processing device divides input audio signals into buffers each containing a number of samples, the buffers overlapping one another. The audio processing device transforms each buffer from the time domain into the frequency domain and implements de-essing as a multi-band compressor that only acts on a designated sibilance band. The audio processing device determines an amount of attenuation in the sibilance band based on comparison of energy level in sibilance band of a buffer to broadband energy level in a previous buffer. The amount of attenuation is also determined based on a zero-crossing rate, as well as a slope and onset of a compression curve.

Abstract: An electronic device and method are disclosed. The electronic device includes an internal microphone, a communication module, and at least one processor. The processor implements the method, including: receiving, through the communication module, a first audio signal input through an external microphone included in an external electronic device communicatively connected to the electronic device, activating the internal microphone in response to detecting a device switch event switching from the external microphone to the internal microphone while receiving the first audio signal, receiving a second audio signal input through the internal microphone, synchronizing and mixing the first audio signal and the second audio signal during a designated first time period, and deactivating the external microphone upon detecting lapse of the designated first time period.

Abstract: A software-based conferencing platform is provided. The platform comprises a plurality of audio sources providing input audio signals, the audio sources including a virtual audio device driver configured to receive far-end input audio signals from a conferencing software module, and a network audio library configured to receive near-end input audio signals from one or more near-end audio devices. The platform further comprises a digital signal processing component configured to receive the input audio signals from the audio sources and generate audio output signals based the received signals, the digital signal processing component comprising an acoustic echo cancellation module configured to apply acoustic echo cancellation techniques to one or more of the near-end input audio signals.

Abstract: A sound source separation filter information estimation device (10) estimates a covariance matrix having information on a correlation between sound source spectra and information on a correlation between channels as information on sound source separation filter information for separating an individual sound source signal from a mixed acoustic signal.

Abstract: To sufficiently suppress noise and reverberation, a convolutional beamformer for calculating, at each time point, a weighted sum of a current signal and a past signal sequence having a predetermined delay and a length of 0 or more such that it increases a probability expressing a speech-likeness of an estimation signals based on a predetermined probability model is acquired where the estimation signals are acquired by applying the convolutional beamformer to frequency-divided observation signals corresponding respectively to a plurality of frequency bands of observation signals acquired by picking up acoustic signals emitted from a sound source, whereupon target signals are acquired by applying the acquired convolutional beamformer to the frequency-divided observation signals.

Abstract: In some embodiments, an echo cancellation method which includes adaptation of at least one prediction filter, with adaptation step size controlled using gradient descent on a set of filter coefficients of the filter, where control of the adaptation step size is based at least in part on a direction of adaptation and a predictability of a gradient of adaptation (e.g., a gradient vector). Other aspects of embodiments of the invention include systems, methods, and computer program products for controlling adaptation step size of adaptive (e.g., low-complexity adaptive) echo cancellation. In some embodiments, adaptation step size control is based on a normalized, scaled gradient of adaptation, or includes smoothing of a normalized gradient of adaptation.

Abstract: A method performed by a processor of a computer system including a headset that is to be worn on a head of a user. The method drives a speaker of the headset with an input audio signal to output sound into an environment. The method determines that the speaker is at least partially covered with a cupped hand. In response to determining that the speaker is at least partially covered with the cupped hand, applying a gain to the input audio signal to reduce an output sound level of the speaker.

Abstract: Methods and apparatus for a communication system having microphones and loudspeakers to determine a noise and speech level estimate for a transformed signal, determine a SNR from the noise and speech level estimates, and determine a gain for the transformed signal to achieve a selected SNR range at a given position. In one embodiment, the gain is determined by adapting an actual gain to follow a target gain, wherein the target gain is adjusted to achieve the selected SNR range.

Abstract: One disclosed example method includes a device receiving an audio signal recorded in a physical environment and applying a de-noise and de-reverberation model onto the audio signal to generate a cleaned audio signal. The de-noise and de-reverberation model is configured to remove noise and reverberation from the audio signal and is trained via a training process. The training process includes training the de-noise and de-reverberation model based on a trained de-noise teacher model and a trained de-reverberation teacher model. The training includes adjusting a portion of parameters of the de-noise and de-reverberation model based on values generated by the de-noise teacher model and the de-reverberation teacher model and then adjusting the parameters of the de-noise and de-reverberation model independently of the de-noise teacher model and the de-reverberation teacher model.

Abstract: Decorrelating an input signal includes allpass filtering to phase shift the first input signal by a phase shift, the allpass filtering comprising filtering with one or more subsequent controllable allpass filter stages, each controllable allpass filter stage having a filter quality and a cut-off frequency. Decorrelating further includes controlling at least one of the filter quality and the cut-off frequency of the controllable allpass filter stages to change over time.

Abstract: A howling detector is described that is configured to receive an input signal and to determine measure of the linearity of a logarithmic representation of the energy of the input signal. In some examples, this triggers gain adjustment (e.g. of a noise control unit) and, in some further examples, the amount of the gain adjustment may be based on an estimation of the maximum stable gain of a noise control unit.

Abstract: This disclosure describes a ceiling tile microphone system that includes a plurality of microphones coupled together as a microphone array and used for beamforming processing, one or more separate processing devices that couple to the microphone array, where one or more separate processing devices further include beamforming, acoustic echo cancellation, and adaptive acoustic processing; a single ceiling tile with an outer surface on the front side of the ceiling tile where the outer surface is acoustically transparent, the microphone array combines with the ceiling tile as a single unit, the ceiling tile being mountable in a drop ceiling in place of a ceiling tile included in the drop ceiling; where the system is used in a drop ceiling mounting configuration; where the microphone array couples to the back side of the ceiling tile and all or part of the system is in the drop space of the drop ceiling.

Abstract: The present embodiments generally relate to enabling participants in an online gathering with networked audio to use a cancelling auralizer at their respective locations to create a common acoustic space or set of acoustic spaces shared among subgroups of participants. For example, there are a set of network connected nodes, and the nodes can contain speakers and microphones, as well as participants and node mixing-processing blocks. The node mixing-processing blocks generate and manipulate signals for playback over the node loudspeakers and for distribution to and from the network. This processing can include cancellation of loudspeaker signals from the microphone signals and auralization of signals according to control parameters that are developed locally and from the network.

Abstract: A fitting agent for a hearing device and related method is disclosed, wherein the fitting agent is configured to initialize a user model comprising a user preference function; obtain a primary test setting for the hearing device; obtain a secondary test setting for the hearing device; present the primary test setting and the secondary test setting to a user; detect a user input of a preferred test setting indicative of a preference for either the primary test setting or the secondary test setting; and update the user model based on hearing device parameters of the preferred test setting, wherein to obtain the secondary test setting comprises: obtain a candidate set of candidate test settings; determine an uncertainty parameter for each candidate test setting; and select the secondary test setting from the candidate set of candidate test settings based on the uncertainty parameters of the candidate test settings.

Abstract: Systems and methods for enhancing a headset user's own voice include at least two outside microphones, an inside microphone, audio input components operable to receive and process the microphone signals, a voice activity detector operable to detect speech presence and absence in the received and/or processed signals, and a cross-over module configured to generate an enhanced voice signal. The audio processing components includes a low frequency branch comprising low pass filter banks, a low frequency spatial filter, a low frequency spectral filter and an equalizer, and a high frequency branch comprising highpass filter banks, a high frequency spatial filter, and a high frequency spectral filter.

Abstract: A method for preparing reference signals for an echo cancellation system disposed in a vehicle, comprising the steps of: receiving a plurality of drive signals, each drive signal being provided to an associated transducer of a plurality of acoustic transducers such that the associated acoustic transducer transduces the drive signal into an acoustic signal, filtering each drive signal with a respective filter of a plurality of filters to produce a plurality of filtered signals, wherein each of the plurality of filters approximates a transfer function from an associated acoustic transducer to a microphone disposed within the vehicle such that the plurality of filtered signals each estimate a respective acoustic signal at the microphone; summing together at least a subset of the plurality of filtered signals to produce a summed reference signal; and outputting the summed reference signal to an echo cancellation system.

Abstract: A method of echo path delay destination and echo cancellation is described in this disclosure. The method includes: obtaining a reference signal, a microphone signal, and a trained multi-task deep neural network, wherein the multi-task deep neural network comprises a first neural network and a second neural network; generating, using the first neural network of the multi-task deep neural network, an estimated echo path delay based on the reference signal and the microphone signal; updating the reference signal based on the estimated echo path delay; and generating, using the second neural network of the multi-task deep neural network, an enhanced microphone signal based on the microphone signal and the updated reference signal.

Abstract: The present application discloses a method for determining a delay between signals, an apparatus, a device and a storage medium, and relates to voice technology. In the method, apparatus, device, and storage medium provided by the present disclosure, by performing down-sampling processing on the signals, the amount of calculation for determining the delay can be reduced, thereby improving the determination efficiency. Moreover, signal segments including alignment positions of the two signals can be estimated in the signal through a currently determined delay, and then the processing can be performed again on the signal segments. In this way, a range for determination can be gradually reduced, that is, an accurate delay can be obtained by just processing shorter signals, which not only ensures the accuracy of the determination, but also reduces the amount of data processing.

Abstract: A loudspeaker is driven with a loudspeaker signal to generate sound, and sound is converted to one or more microphone signals with one or more microphones. The microphone signals are concurrently transformed into far-field beam signals and near-field beam signals. The far-field beam signals and the near-field beam signals are concurrently processed to produce one or more far-field output signals and one or more near-field output signals, respectively. Echo is detected and canceled in the far-field beam signals and in the near-field beam signals. When the echo is not detected above a threshold, the one or more far-field output signals are outputted. When the echo is detected above the threshold, the one or more near-field output signals are outputted. A signal based on the one or more output signals is transmitted.

Abstract: A microphone-loudspeaker integrated apparatus includes a plurality of microphones, a loudspeaker, and a synthesis unit. The loudspeaker is disposed between the plurality of microphones. The synthesis unit synthesizes sounds collected from the plurality of microphones. A distance between the plurality of microphones is set so as to fall within a range within which sound at a frequency needed for recognizing voice is collectable and within a range within which a noise amount due to vibration of the loudspeaker is allowable, when voice recognition is performed based on sound obtained through synthesis by the synthesis unit.

Abstract: A conference system is provided that includes a microphone array unit having a plurality of microphone capsules arranged in or on a board mountable on or in a ceiling of a conference room. The microphone array unit has a steerable beam and a maximum detection angle range. The conference system comprises a processing unit which is configured to receive the output signals of the microphone capsules and to steer the beam based on the received output signal of the microphone array unit. The processing unit is configured to control the microphone array to limit the detection angle range to exclude at least one predetermined exclusion sector in which a noise source is located.

Abstract: The present disclosure provides a microphone apparatus. The microphone apparatus may include a microphone and a vibration sensor. The microphone may be configured to receive a first signal including a voice signal and a first vibration signal. The vibration sensor may be configured to receive a second vibration signal. And the microphone and the vibration sensor are configured such that the first vibration signal may be offset with the second vibration signal.

Abstract: A double talk detection method, a double talk detection apparatus and an echo cancellation system are provided. The double talk detection method comprises: determining, according to an energy ratio between a far-end digital voice signal and a near-end digital voice signal, and a frequency coherence value between the near-end digital voice signal and the far-end digital voice signal, whether a near-end speaker's digital voice signal is present in the near-end digital voice signal. The double talk detection method avoids missing detection and false detection, improves the accuracy of double talk detection, cancels the echo in the near-end voice signal thoroughly when applied in the field of echo cancellation, and improves the communication experience of both talk parties.

Abstract: The present disclosure provides a microphone apparatus. The microphone apparatus may include a microphone and a vibration sensor. The microphone may be configured to receive a first signal including a voice signal and a first vibration signal. The vibration sensor may be configured to receive a second vibration signal. And the microphone and the vibration sensor are configured such that the first vibration signal may be offset with the second vibration signal.

Abstract: A method and an apparatus detect an echo delay. If an electronic device plays an audio and a counter is in an on state, an Nth recording data block received after the counter is turned on is acquired, where the Nth recording data block is a currently received recording data block, and N is greater than 1. Matching is performed between the Nth recording data block and a reference data block that is in a buffer, where the recording data block and the reference data block are data blocks with target duration. After the counter is on, the buffer buffers an audio data block played by the electronic device. If the Nth recording data block matches the reference data block in the buffer, a counting value is acquired. The counter performs counting if the recording data block does not match the reference data block. An echo delay is determined based on the target duration and the counting value.

Abstract: A system for reducing noise for a user includes a first detector configured to generate a first noise signal, wherein the first noise signal is a representation of a first noise that is transmitted to the user through a first sound pathway, and a second detector configured to generate a second noise signal, wherein the second noise signal indicates a second noise perceived by the user. The system also includes a processor configured to determine a noise correction signal based on the first noise signal and/or the second noise signal, and a speaker configured to generate a sound for reducing the noise based on the noise correction signal.

Abstract: A data system for handling user data for a hearing aid user and at least one hearing aid (10, 11) having a memory for storing personal settings for alleviating a hearing loss for the hearing aid user includes a remote server (25) accessible by means of an Internet enabled computer device. A personal communication device (13) is Internet enabled and is able to act as a gateway to the Internet for the at least one hearing aid. The user may create a user account on the remote server, enter personal information into the user account and store the entered personal information on the remote server, and enter the gateway information to the user account. The personal communication device and the at least one hearing aid are provided with respective transceivers for establishing a wireless connection under guidance of said application software. The least one hearing aid uploads the personal settings for alleviating the hearing loss via the gateway to the remote server for storing in the user account.

Abstract: A beamformer system includes an in-ear audio device, such as an earbud, that has three microphones. Two microphones may be disposed on an external face of the audio-device; one microphone may be disposed in or near the ear canal of a user. Data from two microphones is phase-adjusted and combined; a reference signal is generated by phase-adjusting and removing data from the two microphones from data from a third microphone. The combined data is filtered using the reference signal to remove any residual echo, and the resulting data may be used for communications, speech processing, or other uses.

Abstract: A method and a device of sustainably updating a coefficient vector of a finite impulse response FIRfilter. The method includes obtaining (21) a time-varying regularization factor used for iteratively updating the coefficient vector of the FIR filter in a case that the coefficient vector of the FIR filter is used for processing a preset signal; updating (22) the coefficient vector of the FIR filter according to the time-varying regularization factor.

Abstract: Methods, apparatus, systems and articles of manufacture are disclosed to improve detection of audio signatures. An example apparatus includes a TDOA determiner to determine a first time difference of arrival for a first audio sensor of a meter and a second audio sensor of the meter, and a second time difference of arrival for the first audio sensor and a third audio sensor of the meter, a TDOA matcher to determine a match by comparing the first time difference of arrival to a first virtual source time difference of arrival and a second virtual source time difference of arrival, responsive to determining that the first time difference of arrival matches the first virtual source time difference of arrival, identify a first virtual source location as the location of a media presentation device, and remove an audio recording of the second audio sensor to reduce a computational burden on the processor.

Abstract: An echo suppression device includes an echo canceller which suppresses a linear echo signal from an input signal acquired by a microphone; a nonlinear echo estimation unit which, by using a nonlinear echo model indicative of a relationship between at least one of a call reception signal to be output to a speaker and the input signal, and a nonlinear echo signal, estimates the nonlinear echo signal included in the input signal from at least one of the call reception signal and the input signal; a nonlinear echo suppression unit which, by using the estimated nonlinear echo signal, suppresses the nonlinear echo signal from an output signal of the echo canceller; and an echo suppressor which suppresses a residual linear echo signal not suppressed by the echo canceller from an output signal of the nonlinear echo suppression unit.

Abstract: A multi-channel acoustic echo cancellation (AEC) system that includes a residual echo suppressor (RES) that dynamically controls an amount of attenuation to reduce distortion of local speech during double-talk conditions. The RES determines when double-talk conditions are present based on an echo return loss enhancement (ERLE) value. When the ERLE value is above a first threshold value but below a second threshold value, the RES reduces an amount of attenuation applied while generating an RES mask to pass local speech without distortion. When the ERLE value is below the first threshold value or above the second threshold value, the RES applies full attenuation while generating the RES mask in order to suppress a residual echo signal. To further improve RES processing, the RES may apply smoothing across time, smoothing across frequencies, or apply extra echo suppression processing to further attenuate the residual echo signal.

Abstract: A signal processor for providing one or more processed audio signals on the basis of one or more input audio signals is configured to estimate coefficients of an autoregressive reverberation model using the input audio signals and the delayed noise-reduced reverberant signals obtained using a noise reduction. The signal processor is configured to provide noise-reduced reverberant signals using the input audio signals and the estimated coefficients of the autoregressive reverberation model. The signal processor is configured to derive noise-reduced and reverberation-reduced output signals using the noise-reduced reverberant signals and the estimated coefficients of the autoregressive reverberation model. A method and a computer program include a similar functionality.

Abstract: An audio playback system may be adaptable to various situations for improved user experience and audio playback quality. For example, earbuds may be both worn by a same user, worn by two different users within audio range of one another, worn by two different users outside of audio range of one another, etc. A second earbud with a secondary microphone operates in a second mode in which it captures audio and encodes it for transmission to the first earbud having a primary microphone. The first earbud with the primary microphone operates in a first mode in which it mixes the audio received from the second earbud with audio received through its own microphone for playback. The first earbud in the first mode may also delay its own microphone stream to compensate for wireless transmission delay and correlate the two audio streams to improve audio quality in case there are sounds that can be picked up by both microphones.

Abstract: Regions of the audio frequency spectrum that are most susceptible to howling are identified or determined. When detected, these approaches shape the frequency spectrum of the audio going to loudspeakers so that only those howling frequencies are suppressed.

Abstract: Systems and methods are described for compensating for inaccurate echo prediction in audio systems. A signal may be received at a microphone of an audio system, the signal including audio rendered using a spatial audio renderer across a multi-channel audio output. A signal may be received from the spatial audio renderer that indicates a change in rendering of audio. The audio system may then determine if there is echo power within the received signal greater than an expected echo power. After the signal from the spatial audio renderer has been received, the echo suppression applied to the received signal may be modified in response to a determination that the echo power is greater than the expected echo power, the echo suppression attenuating pre-selected frequency bands of the received signal.

Abstract: Speech processing system includes a first sound receiving device, a second sound receiving device, a main controller, and an audio processor. First and second sound receiving devices are configured to generate a main voice signal or a secondary voice signal. A first sensing device in first sound receiving device and a second sensing device in second sound receiving device are configured to output a first sensing signal or a second sensing signal based on a sensing result. Main controller controls first sound receiving device to generate main voice signal and controls second sound receiving device to generate secondary voice signal when receiving first sensing signal. Main controller controls second sound receiving device to generate main voice signal and controls first sound receiving device to generate secondary voice signal when receiving second sensing signal. Audio processor is configured to process main and secondary voice signals into an output voice signal.

Abstract: The audio device according to the present disclosure may include a mixer that adjusts the number of channels of an inputted audio signal based on the number of speakers connected, a transmitter that transmits a test audio signal for speaker setup, to at least one speaker among the plurality of speakers, a feedback receiver that receives a signal of the outputted audio, a controller that determines an output time difference between the plurality of speakers, based on the signal of the outputted audio, and a post-processor that adds an output delay signal to the audio signal of at least one channel of a multi-channel audio signal provided to the plurality of speakers so as to synchronize the outputs of the plurality of speakers, based on the determined output time difference.

Abstract: Techniques for simulating a microphone array and generating synthetic audio data to analyze the microphone array geometry. This reduces the development cost of new microphone arrays by enabling an evaluation of performance metrics (False Rejection Rate (FRR), Word Error Rate (WER), etc.) without building device hardware or collecting data. To generate the synthetic audio data, the system performs acoustic modeling to determine a room impulse response associated with a prototype device (e.g., potential microphone array) in a room. The acoustic modeling is based on two parameters—a device response (information about acoustics and geometry of the prototype device) and a room response (information about acoustics and geometry of the room). The device response can be simulated based on the microphone array geometry, and the room response can be determined using a specialized microphone and a plane wave decomposition algorithm.

Abstract: A method for sound collection includes: converting time domain signals with a number of M collected by devices for sound collecting with a number of M into original frequency domain signals with a number of M; performing beam-forming on the M original frequency domain signals at each of preset grid points, to obtain beam-forming frequency domain signals with a number of N in one-to-one correspondence with the preset grid points; determining an average amplitude of frequency components with a number of N corresponding to each of frequency points with a number of K based on the beam-forming frequency domain signals with a number of N, and synthesizing a synthesized frequency domain signal including the frequency points and having an average amplitude as an amplitude at the each of the frequency points with a number of K; and converting the synthesized frequency domain signal into a synthesized time domain signal.

Abstract: A sound collection apparatus for far-field voice includes a multi-channel analog sound receiver configured to convert an obtained sound signal into an electrical signal; a first analog-to-digital converter coupled to the multi-channel analog sound receiver and configured to convert the electrical signal into a digital signal; and an interface controller coupled to the analog-to-digital converter and configured to transmit the digital signal to a control device via a preset interface. Using the above solutions, the technical problems of high hardware cost and unguaranteed performance of existing sound collection devices can be solved, and the technical effects of effectively reducing the hardware cost and the difficulties of development are achieved.

Abstract: A speech signal processing method is performed at a terminal device, including: obtaining a recorded signal and a to-be-output speech signal, the recorded signal including a noise signal and an echo signal; calculating a loop transfer function according to the recorded signal and the speech signal; calculating a power spectrum of the echo signal and a power spectrum of the noise signal according to the recorded signal, the speech signal, and the loop transfer function; calculating a frequency weighted coefficient according to the two power spectra of the echo signal and the noise signal; adjusting a frequency amplitude of the speech signal based on the frequency weighted coefficient; and outputting the adjusted speech signal to a speaker electrically coupled to the terminal device. As such, the frequency amplitude of the speech signal is automatically adjusted according to the relative frequency distribution of a noise signal and the speech signal.

Abstract: Provided is a method for beamforming by using maximum likelihood estimation in a speech recognition apparatus, including: (a) receiving an input signal (Xn,k) at a time frame n and a frequency k where noise is mixed: (b) determining a probability density function for a target signal (Yn,k) obtained by removing the noise from the input signal satisfies a complex generalized Guassian distribution or a complex gamma distribution where an average value is zero in a time-frequency domain; (c) estimating a variance (?n,k) of the target signal so as to maximize log likelihood for the probability density function; (d) estimating a filter (wk) maximizing a cost function so as to maximize the log likelihood for the probability density function; and (e) repeatedly performing the estimation of the steps (c) and (d) until the filter (wk) coverages, and finally acquiring a final filter (wk).