Yippie, we got a new release. Check the release notes.
Notable points for 0.9.x, please add more feedback 
Deprecated: Docs are up and running.
The new docs aren’t up yet, use the 0.8.2 docs in the meantime.
Wow! Chinese model! You are the best! Thanks very much for your work!
Nice~
In fact they are: https://deepspeech.readthedocs.io/en/r0.9/ we had a bug on readthedocs, and v0.9.0 tag cannot be generated. But we could fix it on the r0.9 branch, so I’ve made sure the redirection is there. When we do a dot release, it should fix as well on RTD.
The newly used noise augmentation does improve my accuracy in noisy environments greatly:
Great to hear that. There are some reports on worsening results for clean speech with the newly released English model with noise augmentation. Maybe this English model is better for noisy envoirnments (which was requested a lot) than for clean speech.
In the benchmark I used above, the clean speech result did get better too (90% → 94%). I didn’t expect this, because LibriSpeech WER got worse, so I think this depends very much on the dataset.
I am glad to see to new version released which support the Chinese language.
And you also provide the checkpoint, pbmm and scorer files.
May I ask you to provide the lm_filtered.arpa and alphabet.txt also?
Because I also keep training a model with my city accent. It is use for the reference only. Thank you.
Do I get it right, that DeepSpeech 0.9.1 fixes the documentation issue and there was no change to native client/binaries compared to v0.9.0?
@dkreutz, you absolutely right, there was some problem with readthedocs and the only solution was to push it to 0.9.1 Thanks for the reminder
And we got a contribution for Python 3.9 support, so we might do a 0.9.2 soonish, and how soon will depend on whether we get other bugs reported or not
And 0.9.2 was just released …