Speech-to-Text models for the Bengali Language. Contains acoustic and language models to be used with deepspeech based ASR.
Acoustic model was trained on Large Bengali ASR training dataset (https://www.openslr.org/53/).
Train size: 203067 samples, 199.99 hours
Dev size: 10690 samples, 10.55 hours
Test size: 2000 samples, 1.84 hours
Language model was trained on OSCAR and Bengali portions of English-Bengali parallel corpora available from OPUS (https://opus.nlpl.eu/).
Accuracy: 30.6% Word error rate (WER), 11.0% Character error rate (CER)
Developer: Alp Öktem
Disclaimer: This model is not tested in production and is provided as-is without any warranty.
