diarize

Attribute speakers and split cues in audio call transcripts.

12|14|Updated Feb 25, 2021

Search Tags:#audio transcription #voice attribution #transcript cleanup #Zoom call analysis #shared-mic scenario

Ready to run

npx skills add https://github.com/ethereumclassic/community-calls --skill diarize

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: diarize
Source: https://github.com/ethereumclassic/community-calls/tree/main/.claude/skills/diarize
Command: npx skills add https://github.com/ethereumclassic/community-calls --skill diarize

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

The Skill addresses the challenge of accurately attributing speaker labels in call transcripts, especially in scenarios where multiple voices are incorrectly merged into a single Zoom participant label due to shared mics or rooms.

Core Features & Use Cases

Voice Attribution: Automatically determines who spoke when in a call where multiple voices are merged in the transcript.
Transcript Splitting: Splits cues at speaker changes, ensuring individual voices are correctly identified and attributed.
Use Case: A user can utilize this Skill to fix a transcript from a call where a group shared a Zoom connection, resulting in incorrect speaker labels for individual voices.

Quick Start

Run the diarize skill with the following command: diarize <recording> --vtt transcripts/<id>.vtt --call <id> --out transcripts/<id>.diarized.vtt

Frequently Asked Questions about diarize

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I fix speaker labels in a Zoom transcript when multiple people shared one mic?▼

Voice attribution identifies individual speakers in call recordings by analyzing audio files and original transcripts. It specifically handles shared-mic scenarios where multiple voices are incorrectly merged into a single participant label in the transcript.

What do I need to provide to split cues and attribute speakers in an audio transcript?▼

To attribute speakers and split cues, you need to provide the original audio recording file, the original VTT transcript, and call metadata. The Skill processes these inputs locally without external dependencies to produce a diarized transcript.

Can I run voice attribution locally without external dependencies or GPU hardware?▼

Yes, you can run voice attribution locally using CPU-only processing without requiring any external dependencies or GPU hardware. The Skill operates entirely on-device to clean up transcripts and attribute speakers in shared-mic scenarios.

How do I diarize a VTT transcript from the command line?▼

To diarize a VTT transcript, run the diarize command by passing the audio recording file, the original VTT transcript path, the call ID, and the desired output path for the corrected diarized VTT file.

Why does my call transcript show only one speaker for multiple people in the same room?▼

A call transcript shows only one speaker for multiple people in the same room due to shared-mic audio issues. When a group shares a single Zoom connection, the system merges all voices into one participant label, requiring transcript cleanup to split and attribute them correctly.