An efficient audio streaming method and apparatus includes a client process implemented on a client or local device and a server process implemented on a remote server or server(s). The client process and server process each have speech recognition components and communicate over a network, and together efficiently manage the detection of speech in an audio signal streamed by the local device to the server for speech recognition and potentially further processing at the server. The client process monitors audio input and in a first detection stage, implements endpointing on the local device to determine when speech is detected. The client process may further determine if a “wakeword” is detected, and then the client process opens a connection and begins streaming audio to the server process via the network. The server process receives the speech audio stream and monitors the audio, implementing endpointing in the server process, to determine when to tell the client process to close the connection and stop streaming audio. The client process continues streaming audio to the server until the server process determines disconnect criteria have been met and tells the client process to stop streaming audio.

An information processing device includes a phonetic converting unit, an HMM converting unit, and a searching unit. The phonetic converting unit converts a phonetic symbol sequence into a hidden Markov model (HMM) state sequence in which states of an HMM are aligned. The HMM converting unit converts the HMM state sequence into a score vector sequence indicating the degree of similarity to a specific pronunciation using a similarity matrix defining the similarity between the states of the HMM. The searching unit searches for a path having a better score for the score vector sequence than that of the other paths out of paths included in a search network and outputs a phonetic symbol sequence corresponding to the retrieved path.

An automatic speech recognition system and a method performed by an automatic speech recognition system are provided. The method includes performing at least two passes of speech activity detection on an acoustic utterance uttered by a speaker. The at least two passes include an initial pass and a subsequent pass. The method further includes estimating at least one of feature statistics and transforms for acoustic feature extraction and acoustic modeling based on an output of an initial pass. The method further includes performing automatic speech recognition using an output of the subsequent pass while bypassing an output of the initial pass to recognize the acoustic utterance.

According to some embodiments, an analog processing portion may receive an audio signal from a microphone. The analog processing portion may then convert the audio signal into sub-band signals and estimate an energy statistic value, such as a Signal-to-Noise Ratio (“SNR”) value, for each sub-band signal. A classification element may classify the estimated energy statistic values with analog processing such that a wakeup signal is generated when voice activity is detected. The wakeup signal may be associated with, for example, a battery-powered, always-listening audio application.

One or more non-transitory computer readable media contain program instructions that, when executed, cause one or more processors to: receive first raw data including one or more first data points generated by a first object of a plurality of objects associated with one or more buildings; generate first input timeseries according to the one or more data points; access a database of interconnected smart entities, the smart entities including object entities representing each of the plurality of objects and data entities representing stored data, the smart entities being interconnected by relational objects indicating relationships between the smart entities; identify a first object entity representing the first object from a first identifier in the first input timeseries; identify a first data entity from a first relational object indicating a relationship between the first object entity and the first data entity; and store the first input timeseries in the first data entity.

A system and method performs multilingual wakeword detection by determining a language corresponding to the wakeword. A first wakeword-detection component, which may execute using a digital-signal processor, determines that audio data includes a representation of the wakeword and determines a language corresponding to the wakeword. A second, more accurate wakeword-detection component may then process the audio data using the language to confirm that it includes the representation of the wakeword. The audio data may then be sent to a remote system for further processing.

This disclosure describes techniques and systems for encoding instructions in audio data that, when output on a speaker of a first device in an environment, cause a second device to output content in the environment. In some instances, the audio data has a frequency that is inaudible to users in the environment. Thus, the first device is able to cause the second device to output the content without users in the environment hearing the instructions. In some instances, the first device also outputs content, and the content output by the second device is played at an offset relative to a position of the content output by the first device.

Audio distortion compensation methods to improve accuracy and efficiency of audio content identification are described. The method is also applicable to speech recognition. Methods to detect the interference from speakers and sources, and distortion to audio from environment and devices, are discussed. Additional methods to detect distortion to the content after performing search and correlation are illustrated. The causes of actual distortion at each client are measured and registered and learnt to generate rules for determining likely distortion and interference sources. The learnt rules are applied at the client, and likely distortions that are detected are compensated or heavily distorted sections are ignored at audio level or signature and feature level based on compute resources available. Further methods to subtract the likely distortions in the query at both audio level and after processing at signature and feature level are described.

A method for generating speech animation from an audio signal includes: receiving the audio signal; transforming the received audio signal into frequency-domain audio features; performing neural-network processing on the frequency-domain audio features to recognize phonemes, wherein the neural-network processing is performed using a neural network trained with a phoneme dataset comprising of audio signals with corresponding ground-truth phoneme labels; and generating the speech animation from the recognized phonemes.

A method for detecting an audio signal, the method comprises: obtaining a speech segment and a non-speech segment of an audio signal to be detected, extracting a first audio feature of the speech segment and a second audio feature of the non-speech segment, detecting the first audio feature using a predetermined speech segment detection model to obtain a first detection score, detecting the second audio feature using a predetermined non-speech segment detection model to obtain a second detection score, and determining whether the audio signal belongs to a target audio based on the first detection score and the second detection score.