transcript-fixer

Correct ASR transcription errors using dictionary rules, GLM AI, and SQLite learning.

1.3k|212|Updated Oct 22, 2025
One-click install
npx skills add https://github.com/daymade/claude-code-skills --skill transcript-fixer
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: transcript-fixer
Source: https://github.com/daymade/claude-code-skills/tree/main/transcript-fixer
Command: npx skills add https://github.com/daymade/claude-code-skills --skill transcript-fixer

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires httpx>=0.24.0, filelock>=3.13.0, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill automates cleaning and correcting transcripts from meetings, lectures, and interviews. It reduces manual proofreading by applying dictionary-based rules, AI-powered corrections, and continuous learning to build a personalized Correction Dictionary that gets smarter over time.

Core Features & Use Cases

  • Dictionary-driven corrections: Apply domain-specific fixes instantly to common transcription errors.
  • AI-assisted corrections: Leverage a GLM-based engine to resolve ambiguous terms and improve quality at scale.
  • Learning from corrections: Build a personal correction database that grows as you fix transcripts.
  • Use Case: Process a batch of meeting notes, fix repeated ASR mistakes, and generate clean final transcripts with minimal manual review.

Quick Start

Initialize the system, add 5-10 common corrections for your domain, then run the full pipeline on a transcript file to generate stage1 (dictionary) and stage2 (AI) outputs, plus a diff report.

Frequently Asked Questions about transcript-fixer

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

FAQPage Schema
How do I fix common transcription errors in ASR transcripts?

Transcription error correction uses dictionary-based rules and AI to automatically fix ASR mistakes in meeting notes, lectures, and interviews. This Skill applies domain-specific corrections, context-aware fixes, and continuous learning to generate clean transcripts with minimal manual review.

Can I correct mixed Chinese and English transcripts with domain-specific terminology?

Yes. This Skill handles bilingual content and technical terminology by combining dictionary-driven corrections, AI-powered edits via GLM, and a learning engine that builds per-domain correction dictionaries for homophones and specialized terms.

How does the learning engine improve corrections over time?

The Skill implements a three-stage pipeline with SQLite persistence to store corrections you make. Your personal correction dictionary grows as you fix transcripts, allowing the system to apply learned patterns to future batches automatically.

What's the quickest way to start cleaning up meeting transcripts?

Initialize the system, add 5-10 common corrections for your domain, then run the full pipeline on a transcript file. The Skill generates stage1 dictionary corrections and stage2 AI-assisted corrections, plus a diff report showing all changes.

Do I need to manually review every correction the AI makes?

No. The three-stage pipeline applies dictionary corrections automatically and generates diff reports for AI edits, reducing manual proofreading significantly. You review only flagged or uncertain corrections before finalizing transcripts.

What file formats and dependencies does this require?

The Skill uses httpx (≥0.24.0) for HTTP requests and filelock (≥3.13.0) for concurrent access. It processes transcript files and generates cleaned outputs with structured diffs; specific input formats are configured during initialization.