Beingpax
VoiceInk stalls during the transcription stage
Hello,
VoiceInk looks like a great app, I just tried it but now, but unfortunately it Keeps hanging at "Transcribing" infinitely.
I have to force quite the app. I use VoiceInk v1.57 (157) on macOS 15.6.1 (24G90), and permissions, except screen recording are enabled. app / mac restart did not help.
Looking forward for some advice! Thank you for your time.
→ Solution: If you have an intel mac, as of now (unfortunately), you need to use a model in the cloud. With deepgram, you can get a ton of free credits.
I updated from v 1.4* to 1.5* and notice that the Transcribing step either stalls completely or takes 30+ seconds to complete, where it was 1-3 seconds before the update, hard to say which version broke this since I had a couple of updates in a row but it is slow to the point of unusable at the moment. It seems to be something in the app itself as the models are still fast.
i'm having this same issue. i just paid for the app, and it immediately stops working the next day. any ideas of how to fix? im using parakeet with nothing else enabled.
I'm facing the same issue - the app doesn't work at all. Also paid for the lifetime licence a couple months ago.
I've encountered the same thing a couple of times now, it's always been some apple code optimizing for the neural engine (check if you have anecompilerservice running in task manager) and has been resolved after a while.
Last time it took around 5-10 min for v3-large (on an M1 Ultra), after that it's lightning quick again.
Not saying there's nothing else going on here, but showing a different message while optimising the model might be nice :)
I don't see anecompilerservice in my Activity Monitor and it stalls and never actually transcribes (even for 1 sentence). 😞
Hey everyone, it seems like model optimization is being triggered somehow because the model cache created during the download is being invalidated by the device. Can you send me an email from within the app by clicking on the Feedback and Issues button?
It will copy important information about the app and your device.
Hey,
I sent the email.
Also, I upgraded to 1.5.9.Alright, and tried this version.
Problem still the same.
It actually completed on one of the trials, took well over 5 minutes to transcribe: "Can you transcribe?", and it randomly inserted it into the email I was writing it to you. I could not reproduce.
Same issue here I think --
After my Macbook M2 has been asleep, then the first transcription takes 2-3 minutes to complete, but after that it's instantaneous until the laptop sleeps again.
I sent an email via the app too.
As per @Beingpax:
- I have an intel mac, and local models are not prepped to work on it - a real bummer if working on patchy network.
- On the flip side, it works with cloud based dictation. a. I registered for a free tier DeepGram account and use their model. It works via an API.
Thank you for the help!
I have the same problem here. It worked well at first, but after using it for a month, it started to freeze frequently. I use local recognition.
There should absolutely be a timeout built into the app, so it's better to receive an error message than have the application hang indefinitely while we wait.
Please try out the latest version, 1.60. You may need to re-download the model.
@Beingpax I upgraded and re-downloaded (Parakeet v2), and so far so good! Thank you!
Edit: Scratch that, it's doing the same thing unfortunately. After the laptop sleeps for a while the first transcription takes a long time.
Just to confirm, this is still happening. It's mostly an annoyance, but the first transcription takes 30+ seconds each morning with Parakeet.
If this is caused by model optimization / cache invalidation, is there some way to have that run automatically in the background on some schedule?