Do I need an API key to use this skill?

For Stage 2 and 3 corrections, a GLM API key from BigModel is required. Stage 1 dictionary corrections work locally and for free without an API key.

Why is 'uv' required for installation?

'uv' is an extremely fast Python package manager used here to ensure that all dependencies and virtual environments are handled safely without interfering with your system Python.

Can I use this for non-English transcripts?

Yes, it is specifically optimized for mixed English and Chinese content, as well as domain-specific technical terminology.

How does the automatic pattern detection work?

The skill tracks corrections that appear at least three times with 80% or higher confidence. It then suggests these as permanent additions to your dictionary during the review process.

What is the Transcript Fixer skill?

It is a specialized extension for Claude Code that automates the cleanup of transcription errors using a combination of local dictionary rules and AI-powered analysis.

Transcript Fixer

Name: Transcript Fixer
Author: daymade

bydaymade

•

147

Productivity & Workflow

Cleans speech-to-text transcription errors using a hybrid approach of dictionary-based rules and AI-powered pattern detection.

About

Transcript Fixer is a specialized Claude Code skill designed to automate the correction of Automated Speech Recognition (ASR) and Speech-to-Text (STT) errors in meeting notes, lectures, and interviews. By combining a persistent SQLite-backed dictionary with advanced AI models, it intelligently identifies homophones, technical jargon, and mixed-language (English/Chinese) mistakes. The skill features a unique learning mechanism that detects repeated correction patterns over time, building a personalized, domain-specific knowledge base that increases in accuracy with every use.

Key Features

Persistent SQLite database for managing custom, domain-specific terminology.
Three-stage correction pipeline: Dictionary-based, AI-powered, or a hybrid of both.
Automated pattern detection that learns from repeated manual corrections.
Optimized for technical terminology and mixed Chinese/English content.
Visual HTML diff generation for easy word-level review of changes.
147 GitHub stars

Use Cases

Developing a shared, collaborative correction database for industry-specific jargon.
Refining technical terminology and proper nouns in business meeting minutes.
Cleaning up academic lecture transcripts that suffer from high homophone error rates.

About

Key Features

Persistent SQLite database for managing custom, domain-specific terminology.
Three-stage correction pipeline: Dictionary-based, AI-powered, or a hybrid of both.
Automated pattern detection that learns from repeated manual corrections.
Optimized for technical terminology and mixed Chinese/English content.
Visual HTML diff generation for easy word-level review of changes.
147 GitHub stars

Use Cases

Developing a shared, collaborative correction database for industry-specific jargon.
Refining technical terminology and proper nouns in business meeting minutes.
Cleaning up academic lecture transcripts that suffer from high homophone error rates.