Abstract: Audio signals produced by microphones can be processed to remove echo and reverberation. The processed signals can be mapped to each other with adaptively estimated impulse responses. One or more of the processed signals, one or more of the mapped signals, and one or more of the impulse responses can be fed to an automatic speech recognizer (ASR) having a deep neural network (DNN), to train the DNN or recognize speech in the input audio signals. Other aspects are described and claimed.

Abstract: An acoustic environment aware method for selecting a high quality audio stream during multi-stream speech recognition. A number of input audio streams are processed to determine if a voice trigger is detected, and if so a voice trigger score is calculated for each stream. An acoustic environment measurement is also calculated for each audio stream. The trigger score and acoustic environment measurement are combined for each audio stream, to select as a preferred audio stream the audio stream with the highest combined score. The preferred audio stream is output to an automatic speech recognizer. Other aspects are also described and claimed.

Abstract: Systems and methods for speech recognition system having a speech processor that is trained to recognize speech by considering (1) a raw microphone signal that includes an echo signal and (2) different types of echo information signals from an echo cancellation system (and optionally different types of ambient noise suppression signals from a noise suppressor). The different types of echo information signals may include those used for echo cancelation and those having echo information. The speech recognition system may convert the raw microphone signal and different types of echo information signals (and optional noise suppression signals) into spectral features in the form of a vector, and a concatenator to combine the feature vectors into a total vector (for a period of time) that is used to train the speech processor, and during use of the speech processor to recognize speech.

Abstract: Systems and methods for speech recognition system having a speech processor that is trained to recognize speech by considering (1) a raw microphone signal that includes an echo signal and (2) different types of echo information signals from an echo cancellation system (and optionally different types of ambient noise suppression signals from a noise suppressor). The different types of echo information signals may include those used for echo cancelation and those having echo information. The speech recognition system may convert the raw microphone signal and different types of echo information signals (and optional noise suppression signals) into spectral features in the form of a vector, and a concatenator to combine the feature vectors into a total vector (for a period of time) that is used to train the speech processor, and during use of the speech processor to recognize speech.

Abstract: Method of detecting voice activity starts with by generating probabilistic models that respectively model features of speech dynamically over time. Probabilistic models may model each feature dependent on a past feature and a current state. Features of speech may include a nonstationary signal presence feature, a periodicity feature, and a sparsity feature. Noise suppressor may then perform noise suppression on an acoustic signal to generate a nonstationary signal presence signal and a noise suppressed acoustic signal. An LPC module may then perform residual analysis on the noise suppressed data signal to generate a periodicity signal and a sparsity signal. Inference generator receives the probabilistic models and receives, in real-time, nonstationary signal presence signal, periodicity signal, and sparsity signal. Inference generator may then generate in real time an estimate of voice activity based on the probabilistic models, nonstationary signal presence signal, periodicity signal, and sparsity signal.

Abstract: Method of detecting voice activity starts with by generating probabilistic models that respectively model features of speech dynamically over time. Probabilistic models may model each feature dependent on a past feature and a current state. Features of speech may include a nonstationary signal presence feature, a periodicity feature, and a sparsity feature. Noise suppressor may then perform noise suppression on an acoustic signal to generate a nonstationary signal presence signal and a noise suppressed acoustic signal. An LPC module may then perform residual analysis on the noise suppressed data signal to generate a periodicity signal and a sparsity signal. Inference generator receives the probabilistic models and receives, in real-time, nonstationary signal presence signal, periodicity signal, and sparsity signal. Inference generator may then generate in real time an estimate of voice activity based on the probabilistic models, nonstationary signal presence signal, periodicity signal, and sparsity signal.