Back to results
US 2026/0221135 A1Published application
Semantically conditioned voice activity detection
AssigneeSoundHound AI IP, LLC
InventorVictor Leitman
Life of this patent
Estimating the term…
Abstract
A method includes recognizing words comprised by a first utterance; interpreting the recognized words according to a grammar comprised by a domain; from the interpreting of the recognized words, determining a timeout period for the first utterance based on the domain of the first utterance; detecting end of voice activity in the first utterance; executing an instruction following an amount of time after detecting end of voice activity of the first utterance in response to the amount of time exceeding the timeout period, the executed instruction based at least in part on interpreting the recognized words.
Claims20
A method comprising:
- receiving sound data representing a spoken utterance of a user;
- recognizing, by a speech recognition engine, one or more candidate word sequences corresponding to the spoken utterance;
- interpreting, by a natural language understanding processor, the one or more candidate word sequences to generate one or more semantic interpretations associated with respective domains or intents;
- determining, by the natural language understanding processor and based at least in part on the semantic interpretations, a timeout period for the spoken utterance,
- wherein the timeout period is variable and depends on at least one of:
- (i) a domain associated with the spoken utterance,
- (ii) an intent associated with the spoken utterance, or
- (iii) a probability associated with one or more of the semantic interpretations;
- detecting an end of voice activity in the spoken utterance;
- and
- executing an instruction in response to expiration of the timeout period following the end of voice activity,
- wherein the instruction is based at least in part on the semantic interpretations of the spoken utterance.
9 more elements
A method comprising:
- receiving sound data representing a spoken utterance of a user;
- incrementally recognizing, by a speech recognition engine, portions of the spoken utterance as the spoken utterance is received;
- incrementally interpreting, by a natural language understanding processor, the portions of the spoken utterance to generate one or more evolving semantic interpretations;
- determining, by the natural language understanding processor and during the spoken utterance, whether the spoken utterance corresponds to a complete interpretable input or to a partial input capable of further extension;
- in response to detecting an end of voice activity during the spoken utterance, initiating a timing operation based on at least one of the evolving semantic interpretations;
- detecting additional voice activity prior to completion of the timing operation and, in response, modifying at least one of:
- (i) the evolving semantic interpretations, and
- (ii) the timing operation;
- and
- executing an instruction after completion of the timing operation when no additional voice activity is detected, the instruction being based at least in part on a selected semantic interpretation.
7 more elements
A method comprising:
- receiving sound data representing a spoken utterance of a user;
- incrementally recognizing, by a speech recognition engine, portions of the spoken utterance to generate one or more transcription hypotheses;
- comparing, by a natural language understanding processor, the transcription hypotheses to a plurality of semantic grammars associated with respective domains to generate a plurality of candidate interpretations and corresponding scores;
- selecting, by the natural language understanding processor, an interpretation based on the scores and updating a conversation state based on the selected interpretation;
- updating the conversation state as additional portions of the spoken utterance are recognized and interpreted;
- determining, based at least in part on the updated conversation state, a timeout period for the spoken utterance;
- detecting an end of voice activity in the spoken utterance;
- initiating a timing operation based on the timeout period and resetting the timing operation upon detection of additional voice activity;
- and
- executing an instruction when the timing operation completes without additional voice activity, the instruction being based at least in part on the updated conversation state.
7 more elements
Dependencies are inferred from the claim text.