Parse prefix-detection in a human-machine interface
View Patent ↗A speech-based human-machine interface that parses words spoken to detect a complete parse and, responsive to so detecting, computes a hypothesis as to whether the words are a prefix to another complete parse. The duration of no voice activity period to determine an end of a sentence depends on the prefix hypothesis. The user's typical speech speed profile and a short-term measure of speech speed also scale the period. Speech speed is measured by the time between words, and the period scaling uses a continuously adaptive algorithm. The system uses a longer cut-off period after a system wake-up event but before it detects any voice activity.
1. A method of adjusting a cut-off period for detecting an end of a sentence, the method comprising:
acquiring a user's longer-term average phoneme rate from a user profile;
detecting inter-word pause lengths between a plurality of words spoken by the user;
computing a shorter-term average phoneme rate from the inter-word pause lengths; and
scaling an NVAD cut-off period based on the longer-term average phoneme rate and the shorter-term average phoneme rate.
2. The method of claim 1 , wherein detecting inter-word pause lengths comprise:
identifying a beginning of a sentence; and
parsing speech from the beginning of the sentence according to a natural language grammer to determine whether the speech received so far constitutes a complete parse.
3. The method of claim 1 ,
wherein detecting inter-word pause lengths comprise:
identifying a beginning of a sentence; and
parsing speech from the beginning of the sentence according to a natural language grammer; and
wherein scaling an NVAD cut-off period comprises:
applying a model to the parsing to produce a hypothesis as to whether the speech received so far is a likely prefix to another complete parse;
scaling the NVAD cut-off period longer when the hypothesis indicates the parse is a likely prefix to another complete parse; and
scaling the NVAD cut-off period shorter when the hypothesis indicates the parse is not a likely prefix to another complete parse.
4. The method of claim 1 , wherein scaling an NVAD cut-off period comprises assigning an NVAD cut-off period that is shorter than an NVAD cut-off period for an incomplete parse.
5. The method of claim 1 , further comprising applying a model to the plurality of words to produce a prefix hypothesis numerical score indicating a likelihood of the plurality of words being a complete sentence; and
wherein scaling an NVAD cut-off period comprises scaling the NVAD cutoff period based on the longer-term average phoneme rate, the shorter-term average phoneme, and the prefix hypothesis numerical score.
6. A method of adjusting a cut-off period for detecting an end of a sentence, the method comprising:
identifying a beginning of a sentence including a first plurality of words spoken by a user;
assigning an NVAD cut-off period that is shorter than an NVAD cut-off period for an incomplete parse based on a prefix hypothesis numerical score produced by applying a model;
computing a user speech speed from inter-word pause lengths between the words in the first plurality of words;
applying the model again to the inter-word pause lengths to produce another prefix hypothesis numerical score indicating a likelihood of the received audio being a complete sentence;
scaling the NVAD cut-off period based on the user speech speed and the other prefix hypothesis numerical score; and
parsing a second plurality of words spoken by the user according to a natural language grammar based on the scaled NVAD cut-off period to determine that the first plurality of words and the second plurality of words collectively constitute a complete parse.
7. The method of claim 6 wherein applying the model is performed responsive to the speech received so far constituting a complete parse.