Contact
dpovey@gmail.com
Phone: 425 247 4129
(Daniel Povey)

Librispeech ASR model

A chain system based on TDNN-F recipe with volume and speed perturbation. An i-vector extractor trained on a 200h subset of the data is also included.

Librispeech ASR Chain 1d

Date
2020-02-03
Uploader
Desh Raj
Recipe
egs/librispeech/s5
Kaldi Version
ea6e1b7
Model Type
Speech Recognition, Factored TDNN, Chain
Error Rate
WER 3.76% on test-clean, 8.92% on test-other
Notes
Reported WER is after rescoring with large 4-gram LM (fglarge).