Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

[BUG] FluidMeet: one speaker's segment runs over the other speaker's replies

Open
  • Is something wrong?
#1,154 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
22/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
macos, swift
Domain
audio-video-rtc

Research direction

Start with the session.json for this meeting and check the 1:57 to 2:15 segment against the other speaker's audio track to see where segment boundaries are set. Then find where the transcript is split and where attributionState is assigned, which is probably where the diarization output meets the two tracks. Done means a segment breaks when the other speaker starts and the short unassigned fragments get a speaker. This is speaker-diarization pipeline work, so it needs maintainer guidance before you start.

Written by the indexing model from the issue text.

Description

FluidMeet puts long stretches of one person's speech into a single segment, even while the other person answers in the middle, so the transcript reads out of order.

Here's the first minute of a 38-minute Google Meet call, with times from session.json. I've replaced my own answers with placeholders:

  • 1:57 to 2:15, other person (one segment): "Oh, yeah. Thank you. Where are you based? Here in the US? Oh, I don't hear that every How long have you been Oh, wow. What made you move over"
  • 2:04, me: [answer to "Where are you based?"]
  • 2:10 to 2:21, me: [answer to "How long have you been there?"]

Read in order, the other person asks four questions in a row and then I give all my answers. In the actual call we went back and forth.

Across the whole call:

  • 48 of 323 segments run right over a complete segment from the other person's audio track. Every segment is marked "overlap": "none", including those 48.
  • 99 segments (about 100 words, mostly one word each, like "uh," and "Meet.") have no speaker assigned (attributionState: unassigned).

I expected segments to split when the other person starts talking, so the transcript keeps the back-and-forth, and short fragments to go to the speaker on that audio track.

Setup: FluidVoice 1.6.12-beta.1, macOS 27.0.1, Apple Silicon. Meeting model parakeet-tdt-v2 (backend local-parakeet-tdt-v2-nemotron-3-v1, pipeline version 13), diarization FluidAudio-offline-v1. Two tracks: application audio (Chrome) and microphone (Insta360 Link). Sound out through the MacBook speakers.

Dominant language
Swift
Stars
11.9k
Forks
863
Avg merge
1d 14h
Merged PRs (30d)
28

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from altic-dev/FluidVoice

All issues in altic-dev/FluidVoice

Similar issues

More Swift issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.