GitHub Repository

Multi-source transcript merging inspired by textual criticism — LLM adjudicates multiple Whisper, YouTube captions & external transcripts for higher quality. Includes speaker diarization and summarization.

23 starsPython

Transcribe-Critic – Merge transcript sources for stronger transcript

Name: Transcribe-Critic – Merge transcript sources for stronger transcript
Availability: InStock
Author: ringger

by ringger·Feb 26, 2026·2 points·1 comment

Visit Project View on HN

AI Analysis

●●●BangerBig BrainZero to One

Textual-criticism approach to transcript merging beats single-model Whisper on accuracy alone.

Strengths

•Blind LLM adjudication (no source bias) + wdiff-based alignment prevents cascading hallucination better than ensemble averaging.
•Applies academic textual-criticism principles (Ringger & Lund 2014) to a modern problem, replacing trained classifiers with LLM.
•Handles speaker labels, timestamps, and external structured transcripts—WhisperX covers neither.

Weaknesses

•Requires running 2–3 Whisper models sequentially, making it slower and costlier than single-pass alternatives like WhisperX.
•No comparison benchmarks provided against WhisperX or manual gold-standard transcripts—claims lack empirical validation.

Similar Projects

AI/ML●●Solid

Transcript-critic, Claude Code skill: transcribe and critically analyze

Everything is wrapped into one /transcribe slash command that downloads, converts, runs whisper.cpp locally, and then fires a long-form analysis prompt that produces timestamped summaries, evidence notes, and flagged logical fallacies. Smart, pragmatic tooling for people who already run local ML and use Claude Code, but the manual path edits and dependency burden (whisper.cpp build, specific model, Claude Code CLI) keep it niche rather than broadly breakout.

Niche GemShip It

jftuga

104mo ago

Developer Tools●●Solid