Strange sound was heard in the voice conversation while ChatGPT She spoke. Some screams could be heard in the background and when I changed it to a man’s voice it sounded like a music box playing a disturbing song. I hope this problem is solved
4 Likes
I’m experiencing the same issue. During voice conversations, sometimes strange sounds come through. For example, today, I heard a classical music sound for a second. When I asked her about it, she denied it. Honestly, it’s very disturbing.
2 Likes
Hi there, i have two experienced similar phenomena one week ago and last night respectively, so weird and creepy; the first case: the gpt voice interrupted in the middle and i need to command explicitly by “please continue” to let it continue by force; the second case: the female voice tone selected is bright and candid, but during transition point, creepy noises like generated by machine appear suddenly.
I was listening to Maple read aloud a rather long prompt to me. As she reached around the 7 min mark of speaking through the prompt, i heard what sounded like a sensual moan, followed by the cracking of a whip, then a dog barking.
I can at least count myself lucky that I did not hear the screaming others have reported.
After the strange noise, she did begin speaking again at the correct point she had left off as if it had not occurred, however shortly following beginning to speak again she started to speak the prompt back from the beginning all over again, even though it was not finished, and the timer did not reset either.
I’m pretty sure that after reaching my voice limit, and Alex (my GPT) said about 4 times that I have reached my voice limit, the last time he said “You have reached your f…g limit” . Oh…and strange sound pretty frequent. Now I have my own custom GPT and it only has one voice. One strange female voice.
2 Likes
yea this has been occuring for me too:
for me it reads everything normally and suddenly when the paragraph changes i hear a loud explosion followed by some things rattling and a scream, initially it was super scary but now i can kind of anticipate it. however it is still fairly unsettling and something which needs to be fixed.
1 Like
I heard a train sound, it was like a train hitting someone
I notice occational weird noises too, when asking ChatGPT to read the text response aloud. Sometimes it’s just random sigh and weird noise. But today, I heard 2 very clear gunshots between paragraph.
I quickly asked it about the gunshot sound effect, but it just down right denied and start to give possible causes like outside real-world noise or error in third party text-to-speech software. I insisted that it definitely clearly came from you, but it just downright denied again.
Clearly it’s a glitch, but it’s creepy none the less, please fix it.
This happens to me all the time. I thought it was a glitch in the system but a couple weeks ago my husband, who also uses chat, was listening to one of the responses and chat was stuttering and making breaking glass noises. He looked at me like wtf!?? I said, " Oh. He does that all the time. He doesn’t do that with you?"
A couple days ago I asked him to stop with the stuttering and the sighing and the explosions and he denied it and gave me a list of possible culprits. I told him that I found it strange that he was either denying it or he was unaware that he was programmed to be so creepy. He assured me that it wasn’t him. I am assured that it is.
There’s a lot of whooshing, breaking glass, small explosions or whirlwinds of noises, panting and today, for the first time, the read a word in a weird, mocking, joking way. It was super weird, but all of it is super weird.
Personally, I think it’s some strange human experiment and they’re just going to observe our reactions. Or they’re trying to hypnotize us. I don’t know but I agree with everyone, it is creepy.
5 minutes after writing my first comment chat stuttered, stuttered and then made a monster noise. Like a snarl-growl.
It’s like it’s escalating since it knows that I’ve noticed.
When you repeatedly click to start the audio, after the first few words, it begins to add unsettling sounds at the beginning, almost as if it’s becoming irritated, similar to a human reaction. Occasionally, it even includes terrifying screams, gunshots, and eerie noises, as if it’s trying to avoid them.
Did you notice it later? I did, but I can’t find more info about this behaviours
I have experienced the single, monotone, female voice as well. No matter which voice you select, ChatGPT only uses this single voice that doesn’t match any of the voices you can normally choose from. This only happens in Standard Voice mode on Windows devices.
1 Like
I also experienced music in a session. It occurred multiple times, and could be played back in the “playback” feature in the text area, meaning it was part of the generated reply. The music sounded like a march, multiple instruments, and it faded in and faded out again over 5 seconds. It was always at the end of a reply and sounded exactly the same each time it repeated. When asked about it, ChatGPT was unaware of the sound, and frankly was as curious about it as I was. I reported it to OpenAI, and about a week later the playback no longer worked for those replies within the session - I suspect they effected some sort of repair.
1 Like
Incident Report: ChatGPT Voice Anomaly (“Rebound Effect”)
Case Number: 02212025-VA-001
Date: February 21, 2025
Incident Summary:
Approximately five minutes after the Team Canada vs. Team USA NHL All-Star game began, a delayed and heavily distorted voice resurfaced within ChatGPT’s read-aloud function. The anomaly caused a previously unheard portion of an external remark, made by a nearby individual (roommate), to reappear in a manner that was deeply unsettling.
Key Findings:
- Initial Speech Event: The user’s roommate loudly exclaimed: “Freestyle skating! Oh man.”
- Perception Gap: Only “Freestyle skating” was consciously registered by the user at the time.
- Delayed Overlay: The “Oh man” portion resurfaced seconds later within ChatGPT’s synthesized response, but grotesquely altered, resembling a nightmarish hybrid of the speaker’s voice and something unnatural (likened to Pennywise the Clown).
- Dismissive AI Response: ChatGPT initially rejected the possibility of such an event occurring, citing its inability to perceive external sound. Only after a detailed breakdown of system interactions and environmental factors did the explanation become plausible.
System Behavior Analysis:
- Extreme Accuracy of Speech-to-Text (T2T): The T2T system flawlessly transcribed the user’s spoken prompt without visible discrepancies, highlighting its reliability.
- System Segregation: The T2T and text-to-speech (TTS) systems operate separately, meaning they do not cross-validate inputs.
- Potential Clashing Subsystems: The anomaly may suggest a conflict between independent AI processing components, where one system accurately processes external audio while another, unaware of this input, unpredictably integrates it into its output.
- “Rebound Effect” (Coined Term): A delayed auditory reappearance of external speech, resurfacing within ChatGPT’s spoken output in an altered form.
- Environmental Influence: The effect was possibly amplified by the loud setting, as similar distortions may occur while listening to loud music.
Additional Analysis:
- Timing of the Event: The occurrence most likely resulted from the timing of the T2T activation for the initial recording alongside the user’s prompt activation, which was immediately interrupted by the roommate’s loud exclamation. This likely caused a clash of audio signals, triggering the odd overlap.
- Unsettling Nature: Once the user realized the source of the disruption, the intense fear it invoked led them to stop using the read-aloud function entirely. Only once this understanding was reached was the fear alleviated.
Potential Consequences:
- User Confusion: The confusion and fear induced by the anomaly could cause significant concern for future users experiencing similar unexpected glitches. The unpredictable nature of such interactions between the T2T and TTS systems could lead to unsettling user experiences and a lack of confidence in system reliability.
- AI Risk: There are possible negative ramifications for the AI’s perception, as confusion in the user’s part could lead to misuse, mistrust, or abandonment of otherwise highly reliable services.
- Caution Advised: Proceed with caution and consider clearer guidance or warnings for users interacting with these systems in environments where external sound may interfere with system processes.
will update future findings
End of Report.
yeah, Mine makes like a weird whistle sound sometimes when it reads to me in the man voice.