mirror of
https://github.com/open-webui/open-webui.git
synced 2026-07-12 02:54:03 -05:00
Call feature does speech-to-text of AI response into user's next query (aka listens 100% of the time) #1376
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @working-name on GitHub (Jun 26, 2024).
Bug Report
Description
Bug Summary:
When I try to use the Call function, my second question contains the AI's response in it, but in an odd way, as if voice-to-text is listening during the AI's text-to-speech response. That should not happen, only the human's input should be converted to text.
Some LLMs like llama3 recognize this duplication of its own response, and will try to make a joke like "I see what you did there".
Steps to Reproduce:
Run the docker container, enable voice. Start a conversation about anything, and ask at least 2 things during the call. You'll notice that the AI's response is included in your next question.
Expected Behavior:
I expect the microphone to be off while the AI's text-to-speech is ongoing, aka follow what the UI says.
Actual Behavior:
It seems the speech-to-text is always active, and transcribes the AI's response alongside the next question asked by the user.
Environment
Open WebUI Version: 0.3.5
Ollama (if applicable): 0.1.46
Operating System: Ubuntu 20.04 on WSL2 on Windows 11, via Docker Desktop 26.1.4, containers run via Docker Compose v2.27.1-desktop.1
Browser (if applicable): Brave Version 1.67.119 Chromium: 126.0.6478.114 (Official Build) (arm64)
Reproduction Details
Confirmation:
I have read and followed all the instructions provided in the README.md.Nothing pertinent in README or TROUBLESHOOTINGLogs and Screenshots
Browser Console Logs:
▶ Brave browser console logs
@working-name commented on GitHub (Jun 26, 2024):
Docker Container Logs:
▶ Relevant docker logs
Installation Method
Docker, no
.envfile, so defaults apply:Additional Information
No clue, browser is on macOS... 🤷♂️