by digityny on 12/20/2024, 6:58:31 PM
Hello, I am a paying customer experiencing issues using the record function on the website from my workplace, although it works perfectly fine from home. It seems there is no option to report an issue through a form or contact anyone directly, thus my report here.. I am using Chrome. When I click the blue button to record the microphone icon appears next to the URL, but the blue line with the control period keeps spinning. Please help. my email is alexrocksthis@gmail.com
by basiskarten on 12/17/2024, 12:11:59 AM
Hey Finn, great project and impressive demo. I had some questions that I was able to answer via the FAQ, except for #3. I'll put the first two here just in case somebody else has the same questions:
1) What about languages other than English? Just English is supported for now.
2) Is this running locally? No.
3) Could you imagine offering a one time payment option? Dragon is currently charging 999 € for their latest version but I really like the fact that I actually own the software after paying once.
In any case, good luck with this!
by blueberrychpstx on 12/11/2024, 3:51:17 PM
Haven’t clicked so fast on anything in a while. Excited to use this on desktop
by davispalmie on 12/11/2024, 3:14:52 PM
Command execution and formatting in the evals is nice instead of just tunneling on transcription accuracy. Are the new benchmarks open sourced?
by newgalfix on 12/11/2024, 5:19:26 PM
Pretty impressive accuracy!
by rcefalu10 on 12/11/2024, 3:10:22 PM
How fast is it? Does it process what I am saying in realtime?
by pablopeniche on 12/11/2024, 3:05:03 PM
Exciting!
The desktop experience is the best way to use Aqua yet. It lets you talk easily into any app and comes with:
- Custom dictionary: can have up to 800 words/phrases at a time, no pronunciation tuning required.
- Context Awareness: automatically identifies relevant words and phrases in the active application. This uses system accessibility APIs (not screenshots, like some others) and is heavily processed on device before inference to preserve privacy.
- Command recognition: the system now shows you what it's going to do before it does it. "Deleting…" or "Adding to list…" or "Fixing Spelling…"
We've also spent a ton of time getting better out-of-the-box accuracy. Our core transcription engine is the most accurate real-time system that we know of. We scored 3.2% WER on Librispeech clean, significantly better than the next best real-time system we tested (Google) at 5.5%. We also released a benchmark that tests accuracy & human-friendly formatting which showed that out of the box, Wispr Flow makes 10x as many mistakes as Aqua Voice for emails and technical writing. The full write-up including audio and code is available here (https://withaqua.com/blog/benchmark-nov-2024)
Everyone uses dictation for different reasons - I started in sixth grade (dyslexic) with Dragon Professional and always wanted it to be more than a clunky substitute for the keyboard. Hopefully Aqua Desktop can be that for some of you.
Would love to hear your comments!
-Finn