Pricing is currently somewhat high, until I figure out how much it costs to run Whisper on cloud GPUs. I'll adjust the pricing accordingly, and perhaps switch to per-minute model.
Probably not realistic. On an M1 Pro MBP, Whisper runs far slower than real time. Think on the order of days for a 2 hour recording.
I’ve been doing transcription work for public meetings. Whisper is truly incredible in terms of error rate even in extremely challenging circumstances (obscure acronyms, unusual terms, unusual names, poor recording quality). I was seeing only a few errors per hour; most things that look like errors are in fact accurate representation of humans saying weird things. But I have to run it on my desktop with CUDA enabled. With the medium model it is iirc barely faster than real time. I only have a 1070 so maybe it is better with more modern hardware.
Whisper does also have some slightly strange behavior with silence and very long recordings. I might do a blog post once I’ve got more experience.
There were at least two errors in the video demo, and that was just 15 seconds of audio. “I can take some notes from a meeting” was transcribed as “I can take some notes from meeting”, and “I click stop [recording]” ended up as “And click the stop”.
Me too - I recently ported the model to plain C/C++ and I am now planning to run it on device and see if the performance is any good. Will post an update when/if it works out
Apple will keep up with anything that SOTA, just with a bit of a lag - so just expect they will be better soon if not already
Word of warning from someone who built an SDK that filled in a processing gap that Apple had (6DOF Monocular SLAM)[1] Apple will eventually make your technology obsolete and their version will be way better. See: ARKit
We open sourced it once ARKit came out because there was no way to monetize it further
Whisper is a game changer in terms of accuracy. It makes Zoom, YouTube, Zoom, Office/Azure, Descript, and Otter.ai transcription look like jokes in comparison.
The step change in transcription accuracy here is significant enough to cross an important threshold for usefulness.
I like that this is so focused on quickly initiating the recording, then just emailing the transcript, rather than trying to keep me in the app. That's exactly how I'd like this type of app to function.
Haha this is just a side project - I'm co-founder of Reflect (https://reflect.app) and more likely we'll be integrating this functionality into our note-taking app.
Cool, yeah that sounds like a logical thing to do. Good luck.
I wish whispermemos was available for Android. I have to look into what's available and what the privacy implications are. Would love to use something like that. A Joplin plugin would be awesome.
On Android one can just use the Microphone input from the keyboard. Just tap and instant transcript, in whatever you like (I use a web-based knowledge thing).
Tap Icon on home screen, tap new, tap Microphone, talk for a few minutes, press save.
It going to your email is quite useful for those that have that email be the primary method of collecting work and notes. It's like that old blackberry concept of the universal inbox, everything just goes to one thing, and get's collected there.
For those with that work flow this is really cool app.
On iOS you can define a two-step Shortcut that does this (type or dictate text and then send it by email to a predefined address [0]). I imagine there are similar facilities on Android.
[0] “Ask for Input” followed by “Send Email”, deactivate “Show Compose Sheet” on the latter. Then use the microphone symbol for speech recognition on the initial text input.
This seems great! I’m excited to try it out. It reminds me of Voiceliner which I’ve been using for a while now. Voiceliner does on device transcription & allows for creating hierarchy to notes / ideas with export functions too
Damn, I wish I knew about Vosk (which Voiceliner uses) a few months ago. I was struggling with Sphinx and couldn't get it to have any accuracy. I will definitely check this out.
Let me mention Just Press Record, which performs the transcription on iOS and macOS with Apple’s speech recognition, which is on-device for several languages.
Seconded - one big win with Just Press Record is you can add it as a complication on an Apple Watch, so you can hit a button anywhere you are, leave a voice memo, and it's transcribed and on all your iCloud devices.
I'm trying to resist a "No wireless. Less space than a Nomad. Lame." response, but iOS already does voice-to-text, so it's hard to see what value-add this offers. I think the comment which inspired you was from an Android user? Or at least not an iOS16 user, where the voice-dictation is much improved.
Single-click launch from the lock screen? That's neat, I don't think I can get there with bare iOS.
No real-time feedback as I speak, so that seems like a point to iOS.
The email is a nice touch, but the trade-off is that you have a copy of all of my recordings now.
Finally, it could be that OpenAI Whisper does a much better job than iOS. I have no idea!
It looks so cool! I'd love to have the transcript send to places other than email. I'm thinking of either a chat app like WhatsApp and Telegram or a productivity app like Notion to keep a more organized journal.
You can use the lock screen widget to start recording - allowing you to quickly capture what's on your mind.
This comment was the inspiration behind this: https://news.ycombinator.com/item?id=32928118
Pricing is currently somewhat high, until I figure out how much it costs to run Whisper on cloud GPUs. I'll adjust the pricing accordingly, and perhaps switch to per-minute model.