search

ahmetoner / whisper-asr-webservice

OpenAI Whisper ASR Webservice API
https://ahmetoner.github.io/whisper-asr-webservice
MIT License
1.86k stars 332 forks source link

Output directory #144

Open someghuser opened 8 months ago

someghuser commented 8 months ago

Sometimes it throws network error. Output files and intermediates in /data/output to save time.

/data/output/<jobid>-<timestamp>-transcription-gb-<notes>.srt /data/output/<jobid>-<timestamp>-translation-en-<notes>.srt

ayancey commented 7 months ago

Please post more specific info about this problem.

someghuser commented 7 months ago

It takes a while to process long audio so sometimes I would get a "network error" message and I don't know if it finished the job or it was some other error. So having a default output directory would help recover finished transcripts. Maybe write a logfile there instead if it was an error.

I encountered this both locally and using google compute.

It's been a while since I used whisper so I don't have the error messages anymore.

ayancey commented 7 months ago

Maybe it's a timeout from your HTTP client? If the transcript is finished but errors out for a different reason, I think we should fix that bug rather than add a "temp" directory feature.

Logs could definitely be improved: #100, #155

someghuser commented 7 months ago

Other apps don't treat that as temp directory but output directory because they support queues like comfyui and automatic 1111, which are stable diffusion and not transcription but still. Being able to download later and not have to deal with an expiring download page seems like an improvement is user experience.

I just use a browser and the default html interface so I don't have any design input in the clients.