ASR for classrooms
- Challenges
- babble noise
- room acoustics
- child speech
- Speech-native Vs LLM-centric Modeling
- LLM-Based ASR
- Contextual biasing
- More context about the audio as metadata
- Contextual biasing
- Whisper is good out of box compared to Wav2Vec
- But if have data to fine tune, wave vec could be a better option