Selected Voice Resets to Generic Female Voice in Standard Mode (Web/Desktop)

ChatGPT Bugs
, ,
1

Since the introduction of ChatGPT Tasks, the selected voice in Standard Mode (Web/Desktop) does not persist.

Observed Behavior:

  • Advanced Voice Mode (AVM): Works fine, the selected voice (e.g., Breeze, Juniper, Ember, Spruce, Cove) is used.
  • Exiting AVM (or starting a new conversation in Standard Mode): Voice resets to a generic female voice (same one used in Custom GPTs).
  • This happens across all selectable voices—it is not limited to a specific voice.
  • The mobile ChatGPT app (Android) does not have this issue—selected voices persist in Standard Mode.
  • The issue affects all browsers (Chrome, Edge, Opera) and the Windows desktop app.

What I Tested So Far:

:small_blue_diamond: Used multiple devices (Main PC, second PC, corporate laptop).
:small_blue_diamond: Used a fresh ChatGPT account (issue persists).
:small_blue_diamond: Tested on a separate network (Vodafone Mobile Hotspot, no relation to home network).
:small_blue_diamond: Tried all suggested fixes (Cache clear, DNS flush, re-login, different browsers) → No effect.
:small_blue_diamond: Recorded and sent multiple video proofs to OpenAI Support.

Possible Cause?

  • The Android app handles voice differently (integrates with system phone call infrastructure).
  • The Web/Desktop version processes voice separately → This might be causing the reset to the generic voice.
  • The issue began with ChatGPT Tasks → Possibly related to changes in how voice sessions are handled.

Official OpenAI Support Response (So Far):

  • OpenAI acknowledged the issue and escalated it to their product team.
  • No estimated fix time yet.
  • Suggested posting here to see if others experience the same issue.

Looking for Feedback:

Has anyone else noticed this issue?
If so, what devices and browsers are you using?
If you have found any workarounds, please share!

Thanks!
Torsten

Edit (add-on): as @onestrongwriter pointed out, the Project Voice chat is affected across all platforms, even Android.

4 Likes
2

Yup been happening to me too. I don’t know if it’s because it’s the first prompt in the voice call or if the prompt is complex enough for it to bug over to a female voice, kind of creepy ahah.

I don’t use much of the android phone app now, Windows 11 ChatGPT app. In a built PC.

2 Likes
3

I’m experiencing a similar issue, but in Projects, not in Tasks.

My selected voice works correctly in standard (non-project) chats.

In Project-based conversations, the UI still shows my chosen voice, but the actual voice speaking in Active Voice is different.

However, the read-aloud function in Projects still uses the correct voice.

I don’t have access to the Tasks feature, so I can’t confirm if the behavior is identical, but it sounds very similar.

2 Likes
4

Thank for pointing that out. I updated Support with your input in another mail.

2 Likes
5

This happened to me prior my access to “Tasks”, but possibly at or after the initial rollout. It began mid-session in a Windows standard voice mode session within the ChatGPT app. I couldn’t change the voice despite cycling through all of them. I rebooted my PC, opened OpenAI in a browser, it was still there. I’ve switched through multiple PC’s, have used a VPN with a new OpenAI account on a new box, and it persists. It doesn’t affect Advanced Voice and doesn’t affect Android. I’ve posted a video of it on Vimeo, but am not allowed to post a link in the text here. You can see it at: Vimeo dot com /1060353860
I’ve reported this to OpenAI, sending them all my troubleshooting steps and a demonstration video. It’s been over a month and have not yet received any explanation, simply “we’re looking into it.”

1 Like
6

That sounds familiar, be it the problem or the response.

Maybe it didn’t happen with the Task roll out, but with Projects since, as mentioned before, the voice bug happens there even on android. Maybe we should bring the issue to discord, Reddit and X as well.

At some point developers have to notice

1 Like
8

It’s interesting that you mentioned this has happened on Android, as it has only happened to me on Windows platforms. I can switch to the same account on Android and standard mode will work as expected. I’ve been speculating that, “perhaps it’s my Microsoft account that was used for all my Windows installs”, which would be insane to think OpenAI is tracking that and assigning some rogue voice based on that fingerprint, but I couldn’t absolutely rule it out. If it’s happening on Android, however, that WOULD rule it out.

I’ve done extensive troubleshooting on this issue - more than I care to admit, probably some 20+ hours - and have isolated nearly every possible variable, such as hardware platform, operating system, account, local network, global region, browser vs app, browser type, Windows versions (10/11). My latest test started with a fresh PC, fresh Windows install, fresh (and different) browser install, new OpenAI account created based on a new e-mail address, using a mobile hotspot instead of my local internet provider and then connecting to the internet via a VPN at a new location - everything was brand new and yet the voice persisted.

I recently realized that one thing that has remained consistent throughout all of my tests is my voice. If OpenAI is fingerprinting my voice and then immediately assigning this rogue voice to that account, that could explain why it persists. Otherwise, the only other logical explanation is that this is happening to everyone, everywhere… and yet I’ve only seen three posts (including yours) from users that make mention of it. Which would mean either nobody else cares or they don’t realize they’re supposed to have a choice of voices in standard voice mode.

I’m wondering as to whether you had a chance to listen to the voice in the video I provided link information for? I can’t post the link directly here, but you can piece together the address from this: vimeo dot com /1060353860

I would be curious to know whether we’re hearing the same exact flat voice, or if yours sounds different. If different, how would you describe it? Does it also sound monotone and detached, or does it have a different oddity?

I’ve affectionately dubbed the mysterious voice “Thursday”, imagining her as a cousin of the fictional character Wednesday Addams. A deadpan, existential “goth girl” voice that is better suited for reading Vogon poetry than engaging in thoughtful, inspirational dialogue.

1 Like
9

It only happens in the Projects Folder voice chat. And yes, it’s the exact same voice as in your video.

I feel your pain. I’ve sent two screen recordings to support.

I tested it on three different computers, multiple browsers, various settings, and accounts. I ruled out location-based issues using a VPN. Additionally, I tested it on seven different Android devices (smartphones and tablets from Samsung, Xiaomi, and Huawei).

Same conclusion every time:

  • Android app works fine.
  • Browsers are broken (including browsers on Android).
  • Windows desktop app is also broken.

Only Read Aloud respects the selected voice, while live conversations always default to the wrong one.

I’ve exchanged 14 emails with support in two weeks, but nothing has changed so far.

At one point I wrote them, that for all the bug hunting, I’d deserve at least a lifetime Pro subscription for free, but they kindly ignored my joke. :wink:

2 Likes
10

Hehe, I considered sending them an invoice for my troubleshooting efforts after painstakingly detailing my efforts and submitting a demonstration video, but I’d definitely settle for even an annual subscription to Pro. :grin:

1 Like
11

Seriously, I’d even settle for a repaired voice mode.
But I would take Pro as an apology for suggesting “did you flush your DNS” after providing hard evidence after a whole week into reporting.

1 Like
12

Yeah, I got a “did you clear your cache and cookies?” after half a dozen e-mail detailing testing across multiple systems and browsers, including a brand new install of everything, including the browser. It was clear they didn’t read the steps I had taken. And, absolutely, I would settle for standard voice simply working on Windows again! :smiling_face_with_three_hearts:

13

I just posted the problem on Discord and also tagged OpenAIDevs on X in a post. Time to get their attention somehow, it’s been over a month. :+1:

1 Like
14

I just had time to test out standard voice in a project folder on both Windows and Android. I haven’t really been using project folders much, and usually in text mode, so I hadn’t noticed, but you’re right - the voice is on Android, too, when in a project folder. On Windows, however, voice mode is disabled completely for me (both advanced and standard) inside project folders. I couldn’t test it there, as it didn’t give me a voice icon to enter voice mode. That’s an interesting clue.

15

OpenAI doesn’t seem to care that an important feature is broken. No reply on twitter or discord.

1 Like
16

Yep, and I never get more than a “we’re looking into it” response via support. :slightly_frowning_face: Guess I have to simply be grateful that it still works on Android.

1 Like
17

I’m here with the same issue. It’s preventing me to groups my conversations in projects. It’s so important to have your selected voice. The other female voice is way too slow for me and gives me the creeps. It’s like your brother now has a female voice ? How can they ignore this?

2 Likes
18

+1 having this issue here.

I’m a Pro subscriber, using Legacy Voice mode (not Advanced Voice).
Voice selection works fine on Android.
But on the Windows App, and Chrome, and Firefox, I get this a weird androgenous sounding voice–it’s unusable on Desktop.

Hope this gets sorted out soon.

1 Like
19

Don’t get your hopes up.
Let me quote from the latest support reply:

When bugs or behaviors like this one are reported and escalated to development, they are assessed and looked at on a priority basis. Things that affect the actual functionality of the app take priority to ensure that the base functions of the app work properly. For example, model or access issue. Due to this, our development team doesn’t have a timeline currently for when this will be fixed.