ASR Improvements: New baseline model for speech recognition improves ASR accuracy in all use cases.

Improvements to ASR adaptation: Utterances that are out of domain (ie. no example utterances provided in the model configuration) now have better speech recognition results.