Streaming audio?
#49
by matthen - opened
I am interested in feeding streaming audio into the model, but it looks like the current ConformerEncoder is not causal.
Is that expected?
Is there any change I can make to the config to make the audio representation causal (and then I could fine-tune)? I have tried setting "chunk_se": 2 in the config.json, but that did not seem to work.
Aha if I set chunk_size from -1 to something positive then it works! chunk_se is not actually used in the code
matthen changed discussion status to closed
Could you share your method？
Very appreciate