Training a model at 8khz?

Hi–

I’d like to try training a model using 8khz audio instead of 16khz. Has anyone attempted this?

Does anyone know if I will need to build from source? Or can I merely change the sox conversions parameters from 16000 to 8000? Anything else I should be wary of?

It should be fine if you re-train from scratch. Training with 8kHz on top of our trained 16kHz, I doubt it’s going to provide good results. You should be able to just change the values in the clients and re-use libdeepspeech.so, so that’s some rebuilding, but simpler.

thank you for the quick reply!