Closed lw64 closed 1 month ago
There's a move to plugin format for the voice services, and this should be one of the supported types soon.
Coqui STT would be a straight-forward drop-in replacement for DeepSpeech, because the APIs are nearly identical :D
also - the latest English model from Coqui STT is much more accurate than the old DeepSpeech model
I'm running Coqui STT on my Picroft as described here (as a REST API the same way DeepSpeech is currently integrated into Mycroft). I needed it to quickly work somehow so it might not be the best solution but maybe it is helpful anyway for someone planning to do it right.
@hslr4 maybe you could create a pull request for the integration into mycroft?
Closing Issue since we're archiving the repo
It seems to me, that the coqui-STT project has reached a point, where it can be used as a backend. There are lots of languages available, and the performance is also very good: "it is running in realtime on a raspberry pi 4 core".
It has also the capability of streaming speech recognition, but as far as I know, that is not yet supported/used anywhere else.
I don't know if a server like for the deepspeech backend, or direct usage of coqui-STT's python bindings is better.