:speech_balloon: An On-Premises, Streaming Speech Recognition System
This release contains pretrained models (RNN-T
and LM
) for two languages: en
and de
and comes with a runnable docker image.
The LM
s have not been tuned and alpha
is hardcoded to 0.1
.
Dataset: ~1400h mixed
Parameters: 70M
WER on Validation Set: 23.8
Dataset: ~800h mixed
Parameters: 70M
WER on Validation Set: 37.6
docker pull iceychris/libreasr:1.0.2
docker run -it -p 8080:8080 iceychris/libreasr:1.0.2