131–135 of 135 posts
I’m sorry to self promote again - but: https://whispermemos.com I’m in love with the idea of pressing button on my Lock Screen and getting a perfect transcription in my inbox. Also, just added emoji summarization in email subject, a small visual reminder of what your memo was about. I hope this is useful to someone!
I recently tried Whisper to transcribe our local Seattle Fire Department radio scanner -- unfortunately it was not reliable enough for my use case, e.g. "adult male hit by car" gets transcribed as "don't mail it by car". I imagine future models will allow the user to input some context to disambiguate. Like if I could give it the audio along with the context "Seattle Fire Department and EMS radio traffic", it would b…
Have you tried the --initial_prompt CLI arg? For my use, I put a bunch of industry jargon and names that are commonly misspelled in there and that fixes 1/3 to 1/2 of the errors. I was initially going to use Azure Cognitive Services and train it on a small amount of test data, after Whisper released for free I use Whisper + openai GPT-3 trained to fix the transcription errors by 1) taking a sample of transcripts by W…
Exciting to see. Curious if there was a benefit to using whisper over something like vosk which can transcribe on mobile device pretty decently. Whisper has other interesting functionality but for straight transcription it seems a bit heavy. Still learning about it and putting it through its paces.
https://alphacephei.com/nsh/2022/10/22/whisper.html
In general, Whisper is more accurate but much more resource heavy. Vosk runs on single core while Whisper needs all CPU cores.
Accuracy difference for clean speech between Vosk-small and Whisper tiny is 2-3% absolute, 20% relative. Not sure how important is it, I would claim it is not that critical.
Numbers there are for original Whisper. Whisper.cpp recommended here is actually 10% worse than stock Whisper for speed considerations. Not that simple.
Vosk is streaming design, you get results with minimum latency of 200ms. Whisper requires you to wait for significant amount of time. If you refactor Whisper for lower latency you will loose a lot of accuracy advantage. Latency is very important for interactive applications like assistants.
Whisper is multilingual and has punctuation, that is a clearly a good advantage. It also can use context properly improving for long recordings.
So on mobile Vosk is still a viable option actually as many others mobile-focused engines.
For server based transcription Whisper is certainly better. But not much better than Nvidia Nemo for example. Not that much publicity for the former though.
It should be all local until it needs information from the internet...