rristo/espnet_ebranchformer_et

e-branchformer et

Espnet2 e-branchformer based recipe (https://github.com/espnet/espnet/tree/master/egs2/librispeech_100/asr1) trained Estonian ASR model using ERR2020 dataset

• WER on ERR2020: 9.9
• WER on mozilla commonvoice_11: 20.8

For usage:

• clone this repo (git clone https://huggingface.co/rristo/espnet_ebranchformer_et)
• go to repo (cd espnet_ebranchformer_et)
• build docker image for needed libraries (build.sh or build.bat)
• run docker container (run.sh or run.sh). This mounts current directory
• run notebook example_usage.ipynb for example usage
  • currently expects audio to be in .wav format

Model description

ASR model for Estonian, uses Estonian Public Broadcasting data ERR2020 data (around 340 hours of audio)

Intended uses & limitations

Pretty much a toy model, trained on limited amount of data. Might not work well on data out of domain (especially spontaneous/noisy data).

Training and evaluation data

Trained on ERR2020 data, evaluated on ERR2020 and mozilla commonvoice test data.

Training procedure

Used espnet e-branchformer based recipe (https://github.com/espnet/espnet/tree/master/egs2/librispeech_100/asr1)

Training results

Look into folder exp/images.

Validation results are in exp/RESULTS.md

Framework versions

• espnet2