Android user here, boiled down to a few key points using the transcription feature that works for me, since I too rant for long periods explaining ideas, etc.
There is a window of time in which if the transcriptions don’t fully send, it fails with an internet connection error. I have very unstable internet where I live, with areas of good and bad speeds/reception. I really notice the difference, so I get the feeling it’s a timeout issue here in regards to the length of audio.
For some cursed reason, rotating your screen mid-recording completely wipes the recording and WILL start from 0. I have lost many just walking around the house.
If you just so happen to be using it during peak hours, time and again it will cut your transcription short and end with “thank you! thank you! Thanks for watching and we’ll see you again,” which is funny but very frustrating.
I don’t know about anyone else, but nowadays I can only get about 1-2 minutes to transcribe reliably. Five minutes is no luck, three minutes if I’m lucky. So I tend to just use it to write 1-minute “paragraphs.” It sucks when I’m speaking to it while driving because that’s when my brain juices move.
These are just workarounds I’ve found. Hope you guys have better luck.
@satelerd, could you share the response from the ticket?
I mean there’s NO WAY that the OpenAI dev team AREN’T aware of this right?? If it’s such a universal issue then why is nothing being done about it? I think OAI need to front up about this
Hello,
This issue annoyed me so much that I created a tool and it made my workflow very smooth.
This is a simple, open-source script that lets you press a single key, talk, and get:
Your voice transcribed with Whisper (locally),
The result copied to clipboard,
And (optionally) sent directly to your current ChatGPT chat or a new one.
Audio and transcripts are saved with clean folder structure to avoid any loss of data, and there’s an optional local LLM integration to fix punctuation and name your recordings smartly.
It runs locally, and is ideal for devs or note-takers who want zero friction when thinking out loud. Currently only tested on Linux, but with some effort it should be possible to make it cross-platform (Mac/Windows) or even run on your phone.
I find my issue to be that I am using the voice to text feature on my web browser, and it works for the first [insert random number bc it’s not consistent] times and then stops working but I never know when it won’t work. To fix it, I refresh my page, but it doesn’t help when it didn’t capture the last thing I just said. I never know when it won’t capture so I am trying to get in a habit of testing before talking but sometimes I forget. Would be great if it captured every time.
For me this has been a problem for a while. more complicated messages where I switch between subjects or go back and forth a bit, they fail. Today though it’s happening with way less complicated messages than I’ve had success with on a regular basis.
Bumping – still a major issue, and over the past month or so I’ve been playing with the voice features (voice-to-text transcription / live voice), it’s actually gotten progressively worse… and now the voice-to-text transcription feature is at the point that it’s completely unusable.
I’m having a similar problem with the Advanced Live Voice mode: I’ve noticed lately that if I speak “too long” per message (between 2-5min.), I get no response and the message is lost.
But as it pertains to the voice-to-text transcription feature to which OP is referring: in-app transcription is frustratingly useless right now. When I turn it on and speak, often it doesn’t do anything – it doesn’t transcribe my words, and lately, it even completely bugs out the app UI (several vital buttons disappear, such as the “send” button(!!!), and they only return when I exit the chat / open a different chat / close the app… thus forcing me to start over again). Additionally, in my experience with this feature, the transcription never works if the message is “too long”… and what is considered “too long” is a mystery, but in my case it used to be anything over 3 minutes – though lately, it’s been anything over 20 seconds (and that’s if the feature works at all!).
Currently, the only workaround I’ve found for this frustrating bug is to use a third-party voice recorder app to record my entire message, and then upload it to ChatGPT. But that means I will not have an easily-readable transcription of my own message that I can return to read later, which is my entire reason for wanting to use the voice-to-text transcription feature in the first place…
I really hope OpenAI fix this soon.
Edit: forgot to mention this is my experience on my Android Pixel 6 phone, and I do not have a VPN on.
This is a feature of ChatGPT that greatly interests me, and until now, I hadn’t noticed it was malfunctioning. However, it hasn’t been working for me for several days. Seeing that some people experienced this bug in 2024, is it possible that it won’t be fixed? It’s quite frustrating.
P.S.: I just realized that transcription isn’t working for me, neither through the website nor the Windows app. It’s as if the microphone isn’t being detected. Things are only getting worse.
I’m just realizing this post is a year old. I’ve only been using Chatgpt for a few months. It wasn’t too bad at first but this last week or two has been brutal. 1 minute voice messages don’t go through on a regular basis.
