Providing correct responses to spoken commands issued to voice-activated devices depends on such devices recognizing the commands accurately. General language models that are agnostic to functional domains have lower quality of speech recognition. An automatic speech recognition system described in this disclosure uses a sound-to-domain classifier to determine probabilities that a particular spoken phrase corresponds to particular functional domains. A language model selector picks a combination of domain-specific language models based on the probabilities. Speech recognition is performed using the selected models.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.