Back to results
US 12,614,552 B1Granted
Joint ASR and speaker error correction post-processing
AssigneeAmazon Technologies, Inc.
InventorsSundararajan Srinivasan, Rohit Paturi, Katrin Kirchhoff, Marc Helbing, Sumit Kumar, Xiang Li
Life of this patent
Estimating the term…
Abstract
Techniques for performing speaker error correction are described. In some examples, speaker error correction is a post-processing task aligned predicted words and predicted one or more speaker identities, wherein the post-processing at least includes: performing predicted speaker error correction on the aligned predicted words and predicted one or more speaker identities to generate a first corrected set of speaker error corrections of the one or more speaker identifiers.
Claims20
A computer-implemented method comprising:
- transcribing, by a transcription service of a provider network, the transcription service implemented as transcription service code stored on one or more computer-readable storage media and executed by one or more processors of the provider network, audio according to a request by at least:
- predicting words in the audio using an automatic speech recognition (ASR) model,
- predicting one or more speaker identities in the audio,
- reconciling the predicted one or more speakers with the predicted words to align the predicted words and predicted one or more speaker identities, and
- performing post-processing on the aligned predicted words and predicted one or more speaker identities, wherein the post-processing at least includes:
- performing, by a lexical corrector of the executing transcription service code, lexical-based predicted speaker error correction on the aligned predicted words and the predicted one or more speaker identities to generate a first corrected set of speaker identities;
- predicting, by an audio featurization sub-segmenter of the executing transcription service code, audio features of the audio;
- and
- generating, by a speaker identity corrector of the executing transcription service code, a second corrected set of speaker identities using the predicted audio features and the first corrected set of speaker identities;
- and
- outputting transcribed audio based on the post-processing, the transcribed audio including the predicted words and the second corrected set of speaker identities.
8 more elements
A computer-implemented method comprising:
- transcribing, by a transcription service implemented as transcription service code stored on one or more computer-readable storage media and executed by one or more processors, audio according to a request by at least:
- predicting words in the audio using an automatic speech recognition (ASR) model,
- predicting one or more speaker identities in the audio,
- reconciling the predicted one or more speaker identities with the predicted words to align the predicted words and predicted one or more speaker identities, and
- performing post-processing on the aligned predicted words and predicted one or more speaker identities, wherein the post-processing at least includes:
- performing, by a lexical corrector of the executing transcription service code, lexical-based predicted speaker error correction on the aligned predicted words and the predicted one or more speaker identities to generate a first corrected set of speaker identities,
- predicting, by an audio featurization sub-segmenter of the executing transcription service code, audio features of the audio, and
- generating, by a speaker identity corrector of the executing transcription service code, a second corrected set of speaker identities using the predicted audio features and the first corrected set of speaker identities;
- and
- outputting transcribed audio based on the post-processing, the transcribed audio including the predicted words and the second corrected set of speaker identities.
7 more elements
A system comprising:
- one or more electronic devices to implement a transcription service as transcription service code stored on one or more computer-readable storage media and executed by one or more processors in a multi-tenant provider network, the transcription service code including instructions that upon execution by the one or more processors cause the transcription service to:
- transcribe audio according to a request by at least:
- predicting words in the audio using an automatic speech recognition (ASR) model,
- predicting one or more speaker identities in the audio,
- reconciling the predicted one or more speaker identities with the predicted words to align the predicted words and predicted one or more speaker identities, and
- performing post-processing on the aligned predicted words and predicted one or more speaker identities, wherein the post-processing at least includes:
- performing, by a lexical corrector of the executing transcription service code, lexical-based predicted speaker error correction on the aligned predicted words and the predicted one or more speaker identities to generate a first corrected set of speaker identities;
- predict, by an audio featurization sub-segmenter of the executing transcription service code, audio features of the audio;
- and
- generate, by a speaker identity corrector of the executing transcription service code, a second corrected set of speaker identities using the predicted audio features and the first corrected set of speaker identities;
- and
- output transcribed audio based on the post-processing, the transcribed audio including the predicted words and the second corrected set of speaker identities.
9 more elements
Dependencies are inferred from the claim text.