HackerTrans
TopNewTrendsCommentsPastAskShowJobs

iceychris

no profile record

Submissions

Show HN: LibreASR – An On-Premises, Streaming Speech Recognition System

github.com
233 points·by iceychris·vor 6 Jahren·71 comments

comments

iceychris
·vor 5 Jahren·discuss
I'm using NixOS with i3 as my daily driver, can recommend.
iceychris
·vor 6 Jahren·discuss
Depending on your objective, noisy data might be useful. I'd like LibreASR to also work in noisy environments, so training on data that is noisy should already help a bit with that. But yeah - stammering and delays are present not only in Common Voice but also Tatoeba and YouTube.
iceychris
·vor 6 Jahren·discuss
Audio boards based on ESP32 boards are quite under the radar and have lovely features for just a few bucks. Running LibreASR on a RPi should also be feasible soon.

Thank you for your kind words! :)
iceychris
·vor 6 Jahren·discuss
Data and compute are the largest hurdles. I only have one GPU and training one model takes 3+ days, so I am limited by that. Also, scraping from YouTube takes time and a lot of storage (multiple TBs).

Mozilla Common Voice data is already used for training.
iceychris
·vor 6 Jahren·discuss
I haven't trained on LibriSpeech exclusively, but yes, the perf on LibriSpeech dev is quite bad, around ~60.0 WER. If the poor alignment of yt captions is the issue, maybe concatenating multiple samples helps a bit.
iceychris
·vor 6 Jahren·discuss
Yes, probably. The data I trained on mostly reflects UK and US accents.
iceychris
·vor 6 Jahren·discuss
Hey blackcat! Your project [0] helped me a lot! Pre-training the encoder sounds great, I'll maybe add it in the future.

[0] https://github.com/theblackcat102/Online-Speech-Recognition
iceychris
·vor 6 Jahren·discuss
I have not yet trained a french model. Also, the gif shows Macron speaking to the congress with his english accent [0]

[0] https://www.youtube.com/watch?v=RqUc1h7bZQ4
iceychris
·vor 6 Jahren·discuss
Right, fixed it, thank you :D
iceychris
·vor 6 Jahren·discuss
As I commented above, very poorly. It's still early days.
iceychris
·vor 6 Jahren·discuss
LibriSpeech, Tatoeba, Common Voice and scraped YouTube videos.
iceychris
·vor 6 Jahren·discuss
The upper transcript is YouTube's automatic transcription. Below is the web app transcribing live. And yes, it is actually missing a few words.
iceychris
·vor 6 Jahren·discuss
Hey HN!

I've been working on this for a while now. While there are other on-premise solutions using older models such as DeepSpeech [0], I haven't found a deployable project supporting multiple languages using the recent RNN-T Architecture [1].

Please note that this does not achieve SotA performance. Also, I've only trained it on one GPU so there might be room for improvement.

Edit: Don't expect good performance :D this is still in early stage development. I am looking for contributers :)

[0] https://github.com/mozilla/DeepSpeech

[1] https://arxiv.org/abs/1811.06621