Users that provide spoken input mixed language are common in many geographies and application domains. Automatic speech recognition in such a context requires multiple natural language understanding (NLU) models to be run in parallel and their outputs to be combined. This disclosure describes techniques to improve the performance of such ASR models by the use of a ranking unit for language determination and assessment of whether the voice input makes sense. A response to the query is provided to the user in the language as determined by the ranking unit.

Creative Commons License

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