Feature extraction for anonymized speech recognition
AssigneeToytalk, Inc.
InventorsOren M. Jacob, Martin Reddy, Brian Langner
Life of this patent
Estimating the term…
Abstract
Various of the disclosed embodiments relate to systems and methods for extracting audio information, e.g. a textual description of speech, from a speech recording while retaining the anonymity of the speaker. In certain embodiments, a third party may perform various aspects of the anonymization and speech processing. Certain embodiments facilitate anonymization in compliance with various legislative requirements even when third parties are involved.
Claims20
A computer-implemented method for speech data processing comprising:
- receiving a raw waveform containing speech data of a user;
- providing the raw waveform to a third party processing system;
- directing the third party processing system to perform an operation upon the waveform, the operation making identification of the user from the waveform more difficult;
- and
- receiving a textual depiction of speech from the third party system.
A computer-implemented method for processing speech data comprising:
- receiving a raw waveform containing speech data of a user;
- generating an anonymized intermediate representation of the speech data based on the raw waveform;
- and
- determining a plurality of textual speech data based on the anonymized intermediate representation.
A computer-implemented method for anonymizing speech data comprising:
- receiving a raw waveform containing speech data of a user;
- segmenting the raw waveform into a plurality of intervals;
- calculating a frequency domain representation of each interval;
- applying a post-processing function to smooth the frequency representation;
- extracting features from at least one of the plurality of intervals;
- and
- providing the features for processing by an analysis system.
Dependencies are inferred from the claim text.