Microsoft Technology Licensing, LLC

Xiong Xiao

Zhuo Chen

Takuya Yoshioka

Shixiong Zhang

Eyal Krupka

Hakan Erdogan

Xuedong D Huang

Jinyu Li

Wei Xiong

Adi Diamant

Amit Kumar Agarwal

Lijuan Qin

Dimitrios Basile Dimitriadis

Changliang Liu

Karen Master Ben-Dor

Ido Leichter

Aviv Hurvitz

Moshe David

Lingfeng Wu

Yoni Smolin

Raz Halaly

Ilya Gurvich

Yifan Gong

Fileno A Alleva

Zhong Meng

Yashesh Gaur

Xiaofei Wang

Zhenghao Wang

Huaming Wang

Naoyuki Kanda

The disclosure herein describes using a transcript generation model for generating a transcript from a multi-speaker audio stream. Audio data including overlapping speech of a plurality of speakers is obtained and a set of frame embeddings are generated from audio data frames of the obtained audio data using an audio data encoder. A set of words and channel change (CC) symbols are generated from the set of frame embeddings using a transcript generation model. The CC symbols are included between pairs of adjacent words that are spoken by different people at the same time. The set of words and CC symbols are transformed into a plurality of transcript lines, wherein words of the set of words are sorted into transcript lines based on the CC symbols, and a multi-speaker transcript is generated based on the plurality of transcript lines. The inclusion of CC symbols by the model enables efficient, accurate multi-speaker transcription.

Training and using a transcript generation model on a multi-speaker audio stream

A method for facilitating a remote conference includes receiving a digital video and a computer-readable audio signal. A face recognition machine is operated to recognize a face of a first conference participant in the digital video, and a speech recognition machine is operated to translate the computer-readable audio signal into a first text. An attribution machine attributes the text to the first conference participant. A second computer-readable audio signal is processed similarly, to obtain a second text attributed to a second conference participant. A transcription machine automatically creates a transcript including the first text attributed to the first conference participant and the second text attributed to the second conference participant.

Computerized intelligent assistant for conferences

A system and method include reception of a first plurality of audio signals, generation of a second plurality of beamformed audio signals based on the first plurality of audio signals, each of the second plurality of beamformed audio signals associated with a respective one of a second plurality of beamformer directions, generation of a first TF mask for a first output channel based on the first plurality of audio signals, determination of a first beamformer direction associated with a first target sound source based on the first TF mask, generation of first features based on the first beamformer direction and the first plurality of audio signals, determination of a second TF mask based on the first features, and application of the second TF mask to one of the second plurality of beamformed audio signals associated with the first beamformer direction.

Low-latency speech separation

Computing devices and methods utilizing a joint speaker location/speaker identification neural network are provided. In one example a computing device receives an audio signal of utterances spoken by multiple persons. Magnitude and phase information features are extracted from the signal and inputted into a joint speaker location and speaker identification neural network. The neural network utilizes both the magnitude and phase information features to determine a change in the person speaking. Output comprising the determination of the change is received from the neural network. The output is then used to perform a speaker recognition function, speaker location function, or both.

Speaker recognition/location using neural network

Examples are disclosed that relate to voice identification enrollment. One example provides a method of voice identification enrollment comprising, during a meeting in which two or more human speakers speak at different times, determining whether one or more conditions of a protocol for sampling meeting audio used to establish human speaker voiceprints are satisfied, and in response to determining that the one or more conditions are satisfied, selecting a sample of meeting audio according to the protocol, the sample representing an utterance made by one of the human speakers. The method further comprises establishing, based at least on the sample, a voiceprint of the human speaker.

Voice identification enrollment

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.

Multi-microphone speech separation

Multi-modal speech localization is achieved using image data captured by one or more cameras, and audio data captured by a microphone array. Audio data captured by each microphone of the array is transformed to obtain a frequency domain representation that is discretized in a plurality of frequency intervals. Image data captured by each camera is used to determine a positioning of each human face. Input data is provided to a previously-trained, audio source localization classifier, including: the frequency domain representation of the audio data captured by each microphone, and the positioning of each human face captured by each camera in which the positioning of each human face represents a candidate audio source. An identified audio source is indicated by the classifier based on the input data that is estimated to be the human face from which the audio data originated.

Multi-modal speech localization

Representative embodiments disclose mechanisms to separate and recognize multiple audio sources (e.g., picking out individual speakers) in an environment where they overlap and interfere with each other. The architecture uses a microphone array to spatially separate out the audio signals. The spatially filtered signals are then input into a plurality of separators, so each signal is input into a corresponding signal. The separators use neural networks to separate out audio sources. The separators typically produce multiple output signals for the single input signals. A post selection processor then assesses the separator outputs to pick the signals with the highest quality output. These signals can be used in a variety of systems such as speech recognition, meeting transcription and enhancement, hearing aids, music information retrieval, speech enhancement and so forth.

Multi-channel speech separation

Computing devices and methods utilizing a joint speaker location/speaker identification neural network are provided. In one example a computing device receives a multi-channel audio signal of an utterance spoken by a user. Magnitude and phase information features are extracted from the signal and inputted into a joint speaker location/speaker identification neural network that is trained via utterances from a plurality of persons. A user embedding comprising speaker identification characteristics and location characteristics is received from the neural network and compared to a plurality of enrollment embeddings extracted from the plurality of utterances that are each associated with an identity of a corresponding person. Based at least on the comparisons, the user is matched to an identity of one of the persons, and the identity of the person is outputted.

Growing community of inventors

Bothell, WA, United States of America

Xiong Xiao