Generate publically releasable corpus to train the language model on

[kdavis-mozilla]

Generate a text upon which the language model can be trained and which can be release under the current licensing.

[kdavis-mozilla]

In light of the test results of #1237 the corpus will be librispeach's training data set.

[kmonachopoulos]

Hello,

Did you generate full mozilla vocab.txt ? There is a LS LM here : http://www.openslr.org/11/ but the librispeech-lm-norm.txt.gz contains transcripts in capital letters. Can we re-build LM using this? Does that make a difference ?

Thanks

[kdavis-mozilla]

@kmonachopoulos I've tried building with that and the quality of recognition goes down.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.