Automatic Speech Recognition

ASR is the automatic conversion of speech into text — converting a spoken utterance into a sequence of word hypotheses that match the original transcription as closely as possible.

Current ASR systems using deep neural networks have reached professional human-transcriber performance in clean speech. However, real-world deployment surfaces hard challenges our system is specifically engineered to handle.

Challenges our ASR handles:

• Physical and social variances of speakers
• Environmental and channel distortions
• Room reverberation in far-field ASR
• Code-switched (multilingual) speech
• Training / test domain mismatch
• Children's speech recognition
• Aged people's speech recognition