Abstract: This document relates to separation of audio signals into speaker-specific signals. One example obtains features reflecting mixed speech signals captured by multiple microphones. The features can be input a neural network and masks can be obtained from the neural network. The masks can be applied one or more of the mixed speech signals captured by one or more of the microphones to obtain two or more separate speaker-specific speech signals, which can then be output.

Abstract: This document relates to separation of audio signals into speaker-specific signals. One example obtains features reflecting mixed speech signals captured by multiple microphones. The features can be input a neural network and masks can be obtained from the neural network. The masks can be applied one or more of the mixed speech signals captured by one or more of the microphones to obtain two or more separate speaker-specific speech signals, which can then be output.

Abstract: The technology relates to converting text to speech utilizing recurrent neural networks (RNNs). The recurrent neural networks may be implemented as multiple modules for determining properties of the text. In embodiments, a part-of-speech RNN module, letter-to-sound RNN module, a linguistic prosody tagger RNN module, and a context awareness and semantic mining RNN module may all be utilized. The properties from the RNN modules are processed by a hyper-structure RNN module that determine the phonetic properties of the input text based on the outputs of the other RNN modules. The hyper-structure RNN module may generate a generation sequence that is capable of being converting to audible speech by a speech synthesizer. The generation sequence may also be optimized by a global optimization module prior to being synthesized into audible speech.

Abstract: A speech recognition system adaptively estimates a warping factor used to reduce speaker variability. The warping factor is estimated using a small window (e.g. 100 ms) of speech. The warping factor is adaptively adjusted as more speech is obtained until the warping factor converges or a pre-defined maximum number of adaptation is reached. The speaker may be placed into a group selected from two or more groups based on characteristics that are associated with the speaker's window of speech. Different step sizes may be used within the different groups when estimating the warping factor. VTLN is applied to the speech input using the estimated warping factor. A linear transformation, including a bias term, may also be computed to assist in normalizing the speech along with the application of the VTLN.

Abstract: The technology relates to converting text to speech utilizing recurrent neural networks (RNNs). The recurrent neural networks may be implemented as multiple modules for determining properties of the text. In embodiments, a part-of-speech RNN module, letter-to-sound RNN module, a linguistic prosody tagger RNN module, and a context awareness and semantic mining RNN module may all be utilized. The properties from the RNN modules are processed by a hyper-structure RNN module that determine the phonetic properties of the input text based on the outputs of the other RNN modules. The hyper-structure RNN module may generate a generation sequence that is capable of being converting to audible speech by a speech synthesizer. The generation sequence may also be optimized by a global optimization module prior to being synthesized into audible speech.

Abstract: The technology relates to performing letter-to-sound conversion utilizing recurrent neural networks (RNNs). The RNNs may be implemented as RNN modules for letter-to-sound conversion. The RNN modules receive text input and convert the text to corresponding phonemes. In determining the corresponding phonemes, the RNN modules may analyze the letters of the text and the letters surrounding the text being analyzed. The RNN modules may also analyze the letters of the text in reverse order. The RNN modules may also receive contextual information about the input text. The letter-to-sound conversion may then also be based on the contextual information that is received. The determined phonemes may be utilized to generate synthesized speech from the input text.

Abstract: Processes capable of accepting linguistic input in one or more languages are generated by re-using existing linguistic components associated with a different anchor language, together with machine translation components that translate between the anchor language and the one or more languages. Linguistic input is directed to machine translation components that translate such input from its language into the anchor language. Those existing linguistic components are then utilized to initiate responsive processing and generate output. Optionally, the output is directed through the machine translation components. A language identifier can initially receive linguistic input and identify the language within which such linguistic input is provided to select an appropriate machine translation component.

Abstract: A speech recognition system adaptively estimates a warping factor used to reduce speaker variability. The warping factor is estimated using a small window (e.g. 100 ms) of speech. The warping factor is adaptively adjusted as more speech is obtained until the warping factor converges or a pre-defined maximum number of adaptation is reached. The speaker may be placed into a group selected from two or more groups based on characteristics that are associated with the speaker's window of speech. Different step sizes may be used within the different groups when estimating the warping factor. VTLN is applied to the speech input using the estimated warping factor. A linear transformation, including a bias term, may also be computed to assist in normalizing the speech along with the application of the VTLN.

Abstract: Processes capable of accepting linguistic input in one or more languages are generated by re-using existing linguistic components associated with a different anchor language, together with machine translation components that translate between the anchor language and the one or more languages. Linguistic input is directed to machine translation components that translate such input from its language into the anchor language. Those existing linguistic components are then utilized to initiate responsive processing and generate output. Optionally, the output is directed through the machine translation components. A language identifier can initially receive linguistic input and identify the language within which such linguistic input is provided to select an appropriate machine translation component.

Abstract: An interactive prompt generation and TTS optimization tool with a user-friendly graphical user interface is provided. The tool accepts HTS abstraction or speech recognition processed input from a user to generate an enhanced initial waveform for synthesis. Acoustic features of the waveform are presented to the user with graphical visualizations enabling the user to modify various parameters of the speech synthesis process and listen to modified versions until an acceptable end product is reached.

Abstract: An interactive prompt generation and TTS optimization tool with a user-friendly graphical user interface is provided. The tool accepts HTS abstraction or speech recognition processed input from a user to generate an enhanced initial waveform for synthesis. Acoustic features of the waveform are presented to the user with graphical visualizations enabling the user to modify various parameters of the speech synthesis process and listen to modified versions until an acceptable end product is reached.

Abstract: A method and computer-readable medium use syllable-like units (SLUs) to decode a pronunciation into a phonetic description. The syllable-like units are generally larger than a single phoneme but smaller than a word. The present invention provides a means for defining these syllable-like units and for generating a language model based on these syllable-like units that can be used in the decoding process. As SLUs are longer than phonemes, they contain more acoustic contextual clues and better lexical constraints for speech recognition. Thus, the phoneme accuracy produced from SLU recognition is much better than all-phone sequence recognition.

Abstract: A language model for a language processing system such as a speech recognition system is constructed from training corpus formed from associated characters, word phrases and context cues. A method and apparatus for generating the training corpus used to train the language model and a system or module using such a language model is disclosed.

Abstract: The present invention is directed to a generating alternatives to words indicative of recognized speech. A reference path of recognized words is generated, based upon input speech data. An operator selection input is received and is indicative of a selected portion of the recognized speech, for which alternatives are to be generated. Boundary conditions for alternatives to be generated are calculated based upon bounds of a reference subpath corresponding to the selected portion of the recognized speech. Alternate subpaths satisfying the boundary conditions are constructed from a hypothesis store which corresponds to the input speech data.

Abstract: An unsupervised adaptation method and apparatus are provided that reduce the storage and time requirements associated with adaptation. Under the invention, utterances are converted into feature vectors, which are decoded to produce a transcript and alignment unit boundaries for the utterance. Individual alignment units and the feature vectors associated with those alignment units are then provided to an alignment function, which aligns the feature vectors with the states of each alignment unit. Because the alignment is performed within alignment unit boundaries, fewer feature vectors are used and the time for alignment is reduced. After alignment, the feature vector dimensions aligned to a state are added to dimension sums that are kept for that state. After all the states in an utterance have had their sums updated, the speech signal and the alignment units are deleted. Once sufficient frames of data have been received to perform adaptive training, the acoustic model is adapted.

Abstract: A method and computer-readable medium convert the text of a word and a user's pronunciation of the word into a phonetic description to be added to a speech recognition lexicon. Initially, two possible phonetic descriptions are generated. One phonetic description is formed from the text of the word. The other phonetic description is formed by decoding a speech signal representing the user's pronunciation of the word. Both phonetic descriptions are scored based on their correspondence to the user's pronunciation. The phonetic description with the highest score is then selected for entry in the speech recognition lexicon.

Abstract: A method and computer-readable medium use syllable-like units (SLUs) to decode a pronunciation into a phonetic description. The syllable-like units are generally larger than a single phoneme but smaller than a word. The present invention provides a means for defining these syllable-like units and for generating a language model based on these syllable-like units that can be used in the decoding process. As SLUs are longer than phonemes, they contain more acoustic contextual clues and better lexical constraints for speech recognition. Thus, the phoneme accuracy produced from SLU recognition is much better than all-phone sequence recognition.

Abstract: A language model for a language processing system such as a speech recognition system is formed as a function of associated characters, word phrases and context cues. A method and apparatus for generating the training corpus used to train the language model and a system or module using such a language model is disclosed.

Abstract: A language model for a language processing system such as a speech recognition system is constructed from training corpus formed from associated characters, word phrases and context cues. A method and apparatus for generating the training corpus used to train the language model and a system or module using such a language model is disclosed.

Abstract: An unsupervised adaptation method and apparatus are provided that reduce the storage and time requirements associated with adaptation. Under the invention, utterances are converted into feature vectors, which are decoded to produce a transcript and alignment unit boundaries for the utterance. Individual alignment units and the feature vectors associated with those alignment units are then provided to an alignment function, which aligns the feature vectors with the states of each alignment unit. Because the alignment is performed within alignment unit boundaries, fewer feature vectors are used and the time for alignment is reduced. After alignment, the feature vector dimensions aligned to a state are added to dimension sums that are kept for that state. After all the states in an utterance have had their sums updated, the speech signal and the alignment units are deleted. Once sufficient frames of data have been received to perform adaptive training, the acoustic model is adapted.