公开(公告)号：US20200372897A1

公开(公告)日：2020-11-26

申请号：US16879714

申请日：2020-05-20

Applicant: Google LLC

Inventor： Eric Dean Battenberg ,  Daisy Stanton ,  Russell John Wyatt Skerry-Ryan ,  Soroosh Mariooryad ,  David Teh-hwa Kao ,  Thomas Edward Bagby ,  Sean Matthew Shannon

IPC: G10L13/047 ,  G10L13/10

Abstract: A method for estimating an embedding capacity includes receiving, at a deterministic reference encoder, a reference audio signal, and determining a reference embedding corresponding to the reference audio signal, the reference embedding having a corresponding embedding dimensionality. The method also includes measuring a first reconstruction loss as a function of the corresponding embedding dimensionality of the reference embedding and obtaining a variational embedding from a variational posterior. The variational embedding has a corresponding embedding dimensionality and a specified capacity. The method also includes measuring a second reconstruction loss as a function of the corresponding embedding dimensionality of the variational embedding and estimating a capacity of the reference embedding by comparing the first measured reconstruction loss for the reference embedding relative to the second measured reconstruction loss for the variational embedding having the specified capacity.

公开(公告)号：US20250131273A1

公开(公告)日：2025-04-24

申请号：US18696052

申请日：2023-09-27

Applicant: Google LLC

Inventor： Soroosh Mariooryad ,  Sean Matthew Shannon ,  Thomas Edward Bagby ,  Siyuan Ma ,  David Teh-Hwa Kao ,  Daisy Antonia Stanton ,  Eric Dean Battenberg ,  Russell John Wyatt Skerry-Ryan

IPC: G06N3/088 ,  G06N3/0455

Abstract: Provided is a noisy channel generative model of two sequences, for example text and speech, which enables uncovering the associations between the two modalities when limited paired data is available. To address the intractability of the exact model under a realistic data set-up, example aspects of the present disclosure include a variational inference approximation. To train this variational model with categorical data, a KL encoder loss approach is proposed which has connections to the wake-sleep algorithm.

公开(公告)号：US20230274728A1

公开(公告)日：2023-08-31

申请号：US18314556

申请日：2023-05-09

Applicant: Google LLC

Inventor： Daisy Stanton ,  Eric Dean Battenberg ,  Russell John Wyatt Skerry-Ryan ,  Soroosh Mariooryad ,  David Teh-hwa Kao ,  Thomas Edward Bagby ,  Sean Matthew Shannon

IPC: G10L13/10 ,  G10L25/30 ,  G10L13/04

CPC classification number: G10L13/10 ,  G10L13/04 ,  G10L25/30 ,  G10L13/08

Abstract: A system for generating an output audio signal includes a context encoder, a text-prediction network, and a text-to-speech (TTS) model. The context encoder is configured to receive one or more context features associated with current input text and process the one or more context features to generate a context embedding associated with the current input text. The text-prediction network is configured to process the current input text and the context embedding to predict, as output, a style embedding for the current input text. The style embedding specifies a specific prosody and/or style for synthesizing the current input text into expressive speech. The TTS model is configured to process the current input text and the style embedding to generate an output audio signal of expressive speech of the current input text. The output audio signal has the specific prosody and/or style specified by the style embedding.

公开(公告)号：US20230260504A1

公开(公告)日：2023-08-17

申请号：US18302764

申请日：2023-04-18

Applicant: Google LLC

Inventor： Eric Dean Battenberg ,  Daisy Stanton ,  Russell John Wyatt Skerry-Ryan ,  Soroosh Mariooryad ,  David Teh-hwa Kao ,  Thomas Edward Bagby ,  Sean Matthew Shannon

IPC: G10L13/047 ,  G10L13/10

CPC classification number: G10L13/047 ,  G10L13/10

Abstract: A method for estimating an embedding capacity includes receiving, at a deterministic reference encoder, a reference audio signal, and determining a reference embedding corresponding to the reference audio signal, the reference embedding having a corresponding embedding dimensionality. The method also includes measuring a first reconstruction loss as a function of the corresponding embedding dimensionality of the reference embedding and obtaining a variational embedding from a variational posterior. The variational embedding has a corresponding embedding dimensionality and a specified capacity. The method also includes measuring a second reconstruction loss as a function of the corresponding embedding dimensionality of the variational embedding and estimating a capacity of the reference embedding by comparing the first measured reconstruction loss for the reference embedding relative to the second measured reconstruction loss for the variational embedding having the specified capacity.

公开(公告)号：US11676573B2

公开(公告)日：2023-06-13

申请号：US16931336

申请日：2020-07-16

Applicant: Google LLC

Inventor： Daisy Stanton ,  Eric Dean Battenberg ,  Russell John Wyatt Skerry-Ryan ,  Soroosh Mariooryad ,  David Teh-Hwa Kao ,  Thomas Edward Bagby ,  Sean Matthew Shannon

IPC: G10L13/00 ,  G10L13/08 ,  G10L13/10 ,  G10L25/30 ,  G10L13/04 ,  G10L13/02 ,  G06N3/044

CPC classification number: G10L13/10 ,  G10L13/04 ,  G10L25/30 ,  G06N3/044 ,  G10L13/02 ,  G10L13/08

Abstract: A system for generating an output audio signal includes a context encoder, a text-prediction network, and a text-to-speech (TTS) model. The context encoder is configured to receive one or more context features associated with current input text and process the one or more context features to generate a context embedding associated with the current input text. The text-prediction network is configured to process the current input text and the context embedding to predict, as output, a style embedding for the current input text. The style embedding specifies a specific prosody and/or style for synthesizing the current input text into expressive speech. The TTS model is configured to process the current input text and the style embedding to generate an output audio signal of expressive speech of the current input text. The output audio signal has the specific prosody and/or style specified by the style embedding.

公开(公告)号：US20210035551A1

公开(公告)日：2021-02-04

申请号：US16931336

申请日：2020-07-16

Applicant: Google LLC

Inventor： Daisy Stanton ,  Eric Dean Battenberg ,  Russell John Wyatt Skerry-Ryan ,  Soroosh Mariooryad ,  David Teh-Hwa Kao ,  Thomas Edward Bagby ,  Sean Matthew Shannon

IPC: G10L13/10

Abstract: A system for generating an output audio signal includes a context encoder, a text-prediction network, and a text-to-speech (TTS) model. The context encoder is configured to receive one or more context features associated with current input text and process the one or more context features to generate a context embedding associated with the current input text. The text-prediction network is configured to process the current input text and the context embedding to predict, as output, a style embedding for the current input text. The style embedding specifies a specific prosody and/or style for synthesizing the current input text into expressive speech The TTS model is configured to process the current input text and the style embedding to generate an output audio signal of expressive speech of the current input text. The output audio signal has the specific prosody and/or style specified by the style embedding.