Awesome-claude-skills local-whisper

install

source · Clone the upstream repo

git clone https://github.com/Anna-Pinewood/awesome-claude-skills

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/Anna-Pinewood/awesome-claude-skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/local-whisper" ~/.claude/skills/anna-pinewood-awesome-claude-skills-local-whisper && rm -rf "$T"

manifest: skills/local-whisper/SKILL.md

source content

Local Whisper — Audio Transcription

Transcribe audio files locally using OpenAI Whisper. Runs on CPU (Apple Silicon MPS turned out to be slower in benchmarks).

Prerequisites

Python 3.11 via Homebrew:
```
/opt/homebrew/bin/python3.11
```
ffmpeg:
```
brew install ffmpeg
```

openai-whisper:

/opt/homebrew/bin/pip3.11 install openai-whisper

If any prerequisite is missing, install it before proceeding.

Script location

~/Documents/awesome-claude-skills/skills/local-whisper/transcribe.py

Usage

/opt/homebrew/bin/python3.11 ~/Documents/awesome-claude-skills/skills/local-whisper/transcribe.py <audio-file> [options]

Options

Flag	Description	Default
`--quality`	Use medium model (better accuracy, ~5x slower)	off (base)
`--language`	Language code (ru, en, etc). Auto-detect if omitted	auto
`--json`	Output JSON with timestamps and segments	off

Two modes

Mode	Model	Speed on M3	When to use
default	base (139 MB)	~5s/min	Quick transcription, drafts, getting the gist
`--quality`	medium (1.5 GB)	~25s/min	When accuracy matters — Russian, accents, noisy audio

Medium is significantly better for Russian language — fewer hallucinations, coherent sentences.

Examples

Fast transcription:

/opt/homebrew/bin/python3.11 ~/Documents/awesome-claude-skills/skills/local-whisper/transcribe.py ~/Downloads/recording.wav

High quality:

/opt/homebrew/bin/python3.11 ~/Documents/awesome-claude-skills/skills/local-whisper/transcribe.py ~/Downloads/recording.wav --quality

With language hint and JSON output:

/opt/homebrew/bin/python3.11 ~/Documents/awesome-claude-skills/skills/local-whisper/transcribe.py ~/Downloads/recording.wav --quality --language ru --json

When invoked as a skill

Take the audio file path from the user's argument or message
Run the transcription script via Bash (add
```
--quality
```
if the user asks for better accuracy or the audio is in Russian)
Present the transcribed text to the user
If quality is poor, suggest re-running with
```
--quality
```