Systems and methods detecting a spoken sentence in a speech recognition system are disclosed herein. Speech data is buffered based on an audio signal captured at a computing device operating in an active mode. The speech data is buffered irrespective of whether the speech data comprises a signature word. The buffered speech data is processed to detect a presence of the sentence comprising at least one command and a query for the computing device. Processing the buffered speech data includes detecting the signature word in the buffered speech data, and in response to detecting the signature word in the speech data, initiating detection of the sentence in the buffered speech data.

A method of operating a customer utterance analysis system includes obtaining a subset of utterances from among a first set of utterances. The method includes encoding, by a sentence encoder, the subset of utterances into multi-dimensional vectors. The method includes generating reduced-dimensionality vectors by reducing a dimensionality of the multi-dimensional vectors. Each vector of the reduced-dimensionality vectors corresponds to an utterance from among the subset of utterances. The method includes performing clustering on the reduced-dimensionality vectors. The method includes, based on the clustering performed on the reduced-dimensionality vectors, arranging the subset of utterances into clusters. The method includes obtaining labels for a least two clusters from among the clusters. The method includes generating training data based on the obtained labels. The method includes training a neural network model to predict an intent of an utterance based on the training data.

A method of operating a customer utterance analysis system includes obtaining a subset of utterances from among a first set of utterances. The method includes encoding, by a sentence encoder, the subset of utterances into multi-dimensional vectors. The method includes generating reduced-dimensionality vectors by reducing a dimensionality of the multi-dimensional vectors. Each vector of the reduced-dimensionality vectors corresponds to an utterance from among the subset of utterances. The method includes performing clustering on the reduced-dimensionality vectors. The method includes, based on the clustering performed on the reduced-dimensionality vectors, arranging the subset of utterances into clusters. The method includes obtaining labels for a least two clusters from among the clusters. The method includes generating training data based on the obtained labels. The method includes training a neural network model to predict an intent of an utterance based on the training data.

A system and method are presented for the correction of packet loss in audio in automatic speech recognition (ASR) systems. Packet loss correction, as presented herein, occurs at the recognition stage without modifying any of the acoustic models generated during training. The behavior of the ASR engine in the absence of packet loss is thus not altered. To accomplish this, the actual input signal may be rectified, the recognition scores may be normalized to account for signal errors, and a best-estimate method using information from previous frames and acoustic models may be used to replace the noisy signal.

A system and method are presented for the correction of packet loss in audio in automatic speech recognition (ASR) systems. Packet loss correction, as presented herein, occurs at the recognition stage without modifying any of the acoustic models generated during training. The behavior of the ASR engine in the absence of packet loss is thus not altered. To accomplish this, the actual input signal may be rectified, the recognition scores may be normalized to account for signal errors, and a best-estimate method using information from previous frames and acoustic models may be used to replace the noisy signal.

Methods, systems, and computer readable media for automated transcription model adaptation includes obtaining audio data from a plurality of audio files. The audio data is transcribed to produce at least one audio file transcription which represents a plurality of transcription alternatives for each audio file. Speech analytics are applied to each audio file transcription. A best transcription is selected from the plurality of transcription alternatives for each audio file. Statistics from the selected best transcription are calculated. An adapted model is created from the calculated statistics.

Methods, systems, and computer readable media for automated transcription model adaptation includes obtaining audio data from a plurality of audio files. The audio data is transcribed to produce at least one audio file transcription which represents a plurality of transcription alternatives for each audio file. Speech analytics are applied to each audio file transcription. A best transcription is selected from the plurality of transcription alternatives for each audio file. Statistics from the selected best transcription are calculated. An adapted model is created from the calculated statistics.

In some embodiments, an exemplary inventive system for improving computer speed and accuracy of automatic speech transcription includes at least components of: a computer processor configured to perform: generating a recognition model specification for a plurality of distinct speech-to-text transcription engines; where each distinct speech-to-text transcription engine corresponds to a respective distinct speech recognition model; receiving at least one audio recording representing a speech of a person; segmenting the audio recording into a plurality of audio segments; determining a respective distinct speech-to-text transcription engine to transcribe a respective audio segment; receiving, from the respective transcription engine, a hypothesis for the respective audio segment; accepting the hypothesis to remove a need to submit the respective audio segment to another distinct speech-to-text transcription engine, resulting in the improved computer speed and the accuracy of automatic speech transcription and generating a transcript of the audio recording from respective accepted hypotheses for the plurality of audio segments.

Described herein is a neural network accelerator with a set of neural processing units and an instruction set for execution on the neural processing units. The instruction set is a compact instruction set including various compute and data move instructions for implementing a neural network. Among the compute instructions are an instruction for performing a fused operation comprising sequential computations, one of which involves matrix multiplication, and an instruction for performing an elementwise vector operation. The instructions in the instruction set are highly configurable and can handle data elements of variable size. The instructions also implement a synchronization mechanism that allows asynchronous execution of data move and compute operations across different components of the neural network accelerator as well as between multiple instances of the neural network accelerator.

There is provided a method of speech-to-speech translation including receiving at a mobile device input speech data associated with speech in a first language and converting the input speech data into input text data using a speech-to-text conversion engine (STT engine) onboard the mobile device. The method also includes translating the input text data to form a translated text data using a text-to-text translation engine (TTT engine) onboard the mobile device. The translated text data is associated with a second language. In addition, the method includes converting the translated text data into output speech data using a text-to-speech conversion engine (TTS engine) onboard the mobile device, and outputting at the mobile device a device output based on the output speech data. Mobile devices and computer-readable storage media for speech-to-speech translation are also provided.