How would you design an acoustic model to remain robust against heavy background noise and reverberation for Deepgram transcription?
