Privacy leakage - I keep seeing other user's output - not my voice-to-text?

[tophat1986]

Bug Description

Bug:

First time using Whispering and I set it up with ElevenLabs scribe_v1 API key.
Privacy and Analytics turned off.
I'm getting random text outputs that are NOT my voice inputs. They appear to be coming from other users?

For example, I've already seen in the first 2 minutes of using the app the below text output when pasting/auto inserting into text input fields:

- （遊戲內購買物品聲）欸？喔！
- 00:06.0 - I'm gonna be a mommy now, and you're going to become an auntie too? Yes it is good for me because my brea......

The second one is a message to a mom about being pregnant - I'm a guy and I don't speak chinese, these could not be more distant from my actual inputs. I'm concerned with how this leakage and privacy concern is working in this app?

Expected behavior:

Steps to reproduce:

1. Just did basic setup per instructions. First time user.

Additional context:

Platform

Windows

Affected Component

Transcription

Whispering Version

v7.11.0

Severity

Moderate (workaround available)

Discord Link

No response

Checklist

• I have searched existing issues to ensure this hasn't been reported
• I have included all relevant information above

[braden-w]

Hey @tophat1986, thanks for the issue, and sorry this was your first experience. The short version is that Whispering has no cloud service or shared exchange between users on our side; the only network hop is the provider you selected (ElevenLabs scribe_v1) and anonymous analytics (which you disabled), so there is no path for other users’ transcripts to appear in the app.

This looks a lot like the “short or empty audio hallucination” behavior we’ve seen before. For example, in #569, the model produced random text like www.feyyaz.tv on near‑silence. In #105, empty/short recordings produce random output in Korean, Chinese, or phrases like “www.beadaholique.com” from what appears to be captioning data in the training set.

Could you confirm whether these outputs happen on recordings longer than 4–5 seconds with clear audio? Also, does setting Language to English instead of Auto change anything, and do you see mic levels moving while recording?

Thanks again!