Same here. Was broken since yesterday evening but now working again (using latest version throughout, no updates or even IDE re-opening) I guess they fixed something server side
Hi All,
the issue was clearly server-side. Why I say so? Because I tested old versions from months ago and they all stopped working at the same time. I use AppImages, meaning, containers with all the code the application needs, it is not a tipical installation where you can remove an individual file of the installation and the voice starts to fail, it is a container that can’t be modified, if it worked in the past and now it doesn’t work, the issue is external to the container image. About the voice-processor.js error we see in the DevTools->Network console, I think it can be and old error left there for many versions (all the old versions I tested have it), but it doesn’t seem to be the cause of loosing the voice feature. I tested this morning again, now the voice is back but the voice-processor.js error is still there in the DevTools->Network console, meaning, that file is not really part of the voice processing, or not a mandatory one.
Hey @dato2000, quick update for everyone in the thread.
The voice transcription issue was fixed on our side yesterday. The fix was backend-only, so no client update was needed. A few of you already confirmed it started working again on its own @dato2000, @Matt_Bunce, @uri111, @Angel_Barragan1 which matches what we expected.
Quick note about the voice-processor.jsERR_FILE_NOT_FOUND in DevTools. @Angel_Barragan1 was right, this one is a red herring. It can show up even when voice works fine, so it was not the real cause.
If you’re still seeing the mic record but not transcribe, let me know your Cursor version, OS, and tier, and we’ll take another look.
I had the same: didn’t work yesterday, started working again today, but:
voice input is flaky. It would transcribe while I talk, but suddenly rewrite the entire prompt, leaving out some parts and while I continue talking then go back to the entire prompt. This would happen several times and if I stop the input at the wrong time, major parts would be missing. This happened several times today already (and confused Claude as my prompt didn’t make sense anymore).
Looking at the devtools, I still see a failed network request when starting the voice input, but it then transcribes. Just in this unreliable back-and-forth manner (edit: I just read about the network error being irrelevant).
Hey @h-audeto, glad recording is working again. What you’re describing sounds like a different issue from the original thread, so let’s split it out.
The transcript rewriting itself mid-speech is the streaming STT updating its partial guesses as more audio comes in. Some refinement is normal, but it shouldn’t drop parts of the final prompt. To dig in, a few details would help:
Cursor version, OS, and layout editor vs Glass or Agents Window
Plan tier
Does it happen on every dictation, or only on longer prompts 15+ seconds?
If you can grab a short screen recording that shows the rewrite happening, that’s perfect. If not, copy-paste the intermediate vs final transcript from one example.
The voice-processor.js failed request you see is unrelated, you can ignore it confirmed in post #37.
Sure, here is a screen recording. It has no audio, but you can see how the prompt keeps changing hugely, way beyond the usual cleanup process. And at the end when there is a long time nothing happening, I keep talking. I also just updated Cursor to see if things get better, but no.
@h-audeto, thanks for calling this out separately. The symptom is different from the original bug (no transcription) and it sounds like new behavior, not a leftover effect.
The text jumping back and forth sounds like progressive STT. The transcript updates as the model gets more context, and sometimes it rewrites what was already said. But “losing whole chunks” sounds more like a regression than expected behavior. I’d like to take a closer look.
Can you share:
Cursor version, OS, and plan (Free/Pro/Pro+/Ultra/Business).
Layout: editor or Glass/unifiedAgent (Cursor Settings > General > Layout).
Does this happen in the Agents Window too, or only in the IDE chat?
What mic are you using: built-in, USB, or Bluetooth?
Plan: I’m on a plan through my employer, so I guess Business.
layout: Editor
In agents window, dictation works entirely different: I speak and there is no live transcript, but only after ending the dictation the text would appear as a block (which, to be honest, I don’t like as a UX as I like to see what the AI heard). But it seems to do well with regards to the described problem.
Mic: USB-attached webcam. Dictation used to work smoothly until last week.
so the voice to text is working, but it’s shifting back and forth during the transcriptiong, i.e. text is jumping around, and works are being written like. “ba–” rather than the word “back”. and itseems to be censoring bad words. “sh—”
the text is very broken
“you got this whole- thing- like we- can- like- what happened?” is all it transcribed in the end after a long dialogue.
@deanrie Works for me now, too, but there is a slight annoyance in the UX; if I type in something, I can’t use the mic. I usually add the lines of code as a reference, but then I’m not able to use the mic. I would like to be able to use the mic even after typing a few words.
