Contextual tagging and biasing of grammars inside word lattices

A system comprising:

• one or more computers;
• and
• one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
• receiving audio data of an utterance;
• generating, by the computing device using a sequence-to-sequence neural network, a word lattice that includes multiple candidate transcriptions of the utterance and that includes transcription confidence scores that each reflect a likelihood that a respective candidate transcription is a match for the utterance;
• determining a context of the computing device;
• based on the context of the computing device:
• identifying, by the computing device, grammars that correspond to the multiple candidate transcriptions;
• determining, for each of the multiple candidate transcriptions, grammar confidence scores that reflect a likelihood that a respective grammar is a match for a respective candidate transcription;
• determining a normalization factor to increase a highest grammar confidence score of the grammar confidence scores to a value of about 1;
• and
• normalizing, using the normalization factor, the grammar confidence scores determined for each of the multiple candidate transcriptions;
• based on the transcription confidence scores and the normalized grammar confidence scores, selecting, from among the candidate transcriptions, a candidate transcription;
• and
• providing the selected candidate transcription as a transcription of the utterance.

A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

• receiving audio data of an utterance;
• generating, using a sequence-to-sequence neural network, a word lattice that includes multiple candidate transcriptions of the utterance and that includes transcription confidence scores that each reflect a likelihood that a respective candidate transcription is a match for the utterance;
• determining a context of the computing device;
• based on the context of the computing device:
• identifying grammars that correspond to the multiple candidate transcriptions;
• determining, for each of the multiple candidate transcriptions, grammar confidence scores that reflect a likelihood that a respective grammar is a match for a respective candidate transcription;
• determining a normalization factor to increase a highest grammar confidence score of the grammar confidence scores to a value of about 1;
• and
• normalizing, using the normalization factor, the grammar confidence scores determined for each of the multiple candidate transcriptions;
• based on the transcription confidence scores and the normalized grammar confidence scores, selecting, from among the candidate transcriptions, a candidate transcription;
• and
• providing the selected candidate transcription as a transcription of the utterance.