ChatGPT Voice can understand what is being said, but it does not currently distinguish which person is speaking during a shared or multi-person conversation. I’d like to propose an optional, opt-in speaker-recognition feature that lets users enroll trusted household voices and keeps each speaker’s identity, Memory, and personalization separate.
A simple example:
A parent is talking to ChatGPT and then their child starts speaking. ChatGPT may continue treating everything being said as though it came from the account owner.
I think ChatGPT Voice should optionally allow users to create enrolled speaker profiles such as:
- Me
- Spouse/partner
- Child
- Other household member
Then Voice could internally associate each statement with the appropriate speaker.
Parent: “He’s having trouble keeping up with his assignments.”
Child: “I usually finish them, but I forget to turn them in.”
ChatGPT would know those statements came from two different people.
This could improve:
- Family conversations
- Shared devices
- Parent/child educational use
- Couples using ChatGPT together
- Memory accuracy
- Age-appropriate interactions
- Cars and hands-free use
- Accessibility
- Multi-person brainstorming
- Future household assistants
It could also prevent another person’s statements from accidentally becoming associated with the account owner’s Memory.
This should be completely opt-in. Users should explicitly enroll voices rather than ChatGPT automatically identifying everyone it hears.
Important safeguards could include:
- Explicit voice enrollment
- Speaker consent
- Easy deletion
- “Unknown Speaker” when confidence isn’t high enough
- Separate Memory permissions by speaker
- Additional protections for minors
- On-device voice matching where feasible
A possible setting could look like:
Settings → Voice → Speaker Profiles
Recognize speakers: On / Off
Enrolled:
- Me
- Partner
- Child
Unknown speakers:
- Treat as guest
- Ask who’s speaking
- Don’t use personal Memory
As ChatGPT Voice becomes more conversational, distinguishing who is speaking, not merely understanding what is being said, could become an important part of making it work well as a multi-person assistant.