Modèle Français 0.5.1
Pre-release
Pre-release
Jeux de données :
- Lingua Libre (~40h)
- Common Voice FR (v2) (~490h, en autorisant jusqu'à 32 duplicatas)
- Training Speech (~180h)
- African Accented French (~15h)
- M-AILABS French (~315h)
Total : ~1040h
Paramètres :
EPOCHS=30
LEARNING_RATE=0.0001
DROPOUT=0.3
BATCH_SIZE=64
LM_ALPHA=0.7203202402564637
LM_BETA=1.5747698919871918
Language Model : dump wikipedia + dump débats assemblée nationale.
Fonctionne avec DeepSpeech v0.7
, v0.8
.
Résultats test set:
Test on /mnt/extracted/data/African_Accented_French/African_Accented_French/African_Accented_French_test.csv - WER: 0.442362, CER: 0.235577, loss: 42.941334
Test on /mnt/extracted/data/M-AILABS/fr_FR/fr_FR_test.csv - WER: 0.092794, CER: 0.026505, loss: 11.276774
Test on /mnt/extracted/data/trainingspeech/ts_2019-04-11_fr_FR_test.csv - WER: 0.200373, CER: 0.059958, loss: 16.225618
Test on /mnt/extracted/data/cv-fr/clips/test.csv - WER: 0.300508, CER: 0.147202, loss: 39.204407
Test on /mnt/extracted/data/lingualibre/lingua_libre_Q21-fra-French_test.csv - WER: 0.577170, CER: 0.171211, loss: 6.977585