Online language model interpolation for automatic speech recognition

A computer-implemented method comprising:

• receiving a domain grammar;
• determining an interpolated grammar based on the domain grammar and a base grammar;
• determining a delta domain grammar based on the interpolated grammar and an augmented first grammar stored prior to acquisition of the domain grammar, the augmented first grammar generated based on the base grammar and allowing for insertion of an out-of-vocabulary class;
• determining a first out-of-vocabulary class based on a comparison between the domain grammar and the base grammar;
• inserting the first out-of-vocabulary class into a composed transducer composed of the augmented first grammar and one or more other transducers to generate an updated composed transducer, where the updated composed transducer is stored prior to acquisition of the domain grammar;
• dynamically composing the delta domain grammar and the updated composed transducer;
• and
• performing speech recognition using the dynamic composition of the delta domain grammar and the updated composed transducer.

A system comprising:

• a meeting language model service to:
• generate a domain grammar associated with a meeting;
• and
• a transcription service to;
• receive a domain grammar;
• determine an interpolated grammar based on the domain grammar and a base grammar;
• determine a delta domain grammar based on the interpolated grammar and an augmented first grammar stored prior to acquisition of the domain grammar, the augmented first grammar generated based on the base grammar and allowing for insertion of an out-of-vocabulary class;
• determine a first out-of-vocabulary class based on a comparison between the domain grammar and the base grammar;
• insert the first out-of-vocabulary class into a composed transducer composed of the augmented first grammar and one or more other transducers to generate an updated composed transducer, where the updated composed transducer is composed and stored prior to acquisition of the domain grammar;
• compose the delta domain grammar and the updated composed transducer;
• and
• perform speech recognition on audio of the meeting using the composition of the delta domain grammar and the updated composed transducer.