Visuo-phonetic decoding using multi-stream and context-dependent models for an ultrasound-based silent speech interface.