The cursor IDE has a feature to speak into the chat. However, there is an odd bug, that sometimes makes it so, that nothing is being recognized. The wave form is moving, but no written words appear.
Klicking into the text field, on stop and then on record again does fix it sometimes, but just now I tried around 20 times and nothing worked.
I am on the Pro plan. It is the same with the Teams plan.
I normally use voice a lot, so that appears to happen often. It is quite annoying, especially when I didn’t notice and a lot of talking was wasted.
Hopefully, this can be fixed. If needed, I can create a video if it happens.
Thank you
Steps to Reproduce
It happens randomly, at least it looks like it. Or overload on your side somewhere.
Expected Behavior
At least start transcribing the words at one point in time. Not doing nothing at all forever.
Me too. For some reason most of the time what fixes mine is that I reload/refresh Cursor browser. (I always have the internal browser of my project open.) No idea if that fix works for you, but you can try it.
Developer: Reload Window does sometimes work. But lately it also happens very often, that it stops transcribing in th emid of me speaking. And never goes on.
Which is even worse, since you only can click the listeningbutton once per prompt.
Open new chat and there will be a fresh voice button. Continue speaking where you got cut off. Then cut and paste into the first chat. But mostly opening new chat has the same issue.
Hey @Steph_Nachname sorry about the wasted dictation. Quick workarounds while we sort this out:
If no text shows in the first few seconds, stop and restart the recording.
If it cuts out mid-prompt, @jes’s tip is the one to use: open a new chat for a fresh voice button, finish there, then paste it back. A Developer: Reload Window also helps some.
This is on our side, not anything in your setup, and it’s an issue we’re tracking. If it keeps happening, drop a rough timestamp (with your timezone) and I’ll dig into the logs
Thanks Mohit, it’s very disturbing currently. I basically cannot really use it. It seems to have gotten worse over the last weeks.
2026-09-04 22:26 I could dictate some, then it stopped.
2026-09-04 23:10 Did not start transcribing. Multiple times.
2026-09-05 01:37 Stopped transcribing in the middle.
2026-09-05 01:59 Worked initially, after thinking pause it did not proceed.
2026-09-05 02:19 The same. It seems, as if the listener stops, while still pretends to listen. It would be more honest to make the stop visible. But it still is not ideal. sometimes you have to sort your thoughts and proceed talking, it’s not great to find out afterwards, that you lost your words.
Also, it should be possible to proceed a voice input. Currently it is only posisble if the input window is completely empty.
2026-09-05 03:23 The same.
2026-09-05 03:33 The same.
I guess, that is enough. Now it was more, that it stopped in the middle.
Thanks @Steph_Nachname, and @jes for the workaround. I pulled the sessions around your timestamps and confirmed it is on our side: your audio reaches us fine, but the speech-to-text service intermittently returns nothing (and sometimes cuts out mid-way) while the recorder keeps looking active. That is the issue we are tracking, and it lines up with the times you listed.
Your point stands that the recorder should show when it has stopped, and I have passed along the ask to continue a dictation into a non-empty input box. Same workarounds hold meanwhile: restart if nothing shows in the first few seconds, and for a mid-prompt cutout use @jes’s tip (new chat for a fresh voice button, finish there, paste back). I will post here when there is an update.
Thanks @mohitjain, good to hear, that the issue is being recognized. It is quite troubling for the workflow.
I’m not sure if the workarounds actually work. It sometimes feels as if it is pure luck to get a point in time where the server is not overloaded or something.
It happened multiple times now, that it stopped mid-transcribing and I waited. Then suddenly the whole transcription got deleted and I had to start again.
Also, if the agent is working, normally you can already write the next message. But it does not offer voice recognition, while typing works.