# Speech To Text

> >

- **Type:** Skill
- **Install:** `agentstack add skill-claude-world-claude-agent-speech-to-text`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [claude-world](https://agentstack.voostack.com/s/claude-world)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [claude-world](https://github.com/claude-world)
- **Source:** https://github.com/claude-world/claude-agent/tree/main/.claude/skills/speech-to-text

## Install

```sh
agentstack add skill-claude-world-claude-agent-speech-to-text
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Speech to Text

Transcribe audio files locally using OpenAI Whisper — no API key required.

## Prerequisites

Install local Whisper:
```bash
pip install openai-whisper
# Requires Python 3.8+
```

Also install ffmpeg (required by Whisper for audio processing):
```bash
brew install ffmpeg
```

First run downloads the model automatically to `~/.cache/whisper/`.

Model sizes (speed vs accuracy tradeoff):
| Model | Size | Speed | Accuracy |
|-------|------|-------|----------|
| tiny | 39MB | Very fast | Basic |
| base | 74MB | Fast | Good |
| small | 244MB | Moderate | Better |
| medium | 769MB | Slow | High |
| large | 1.5GB | Very slow | Best |

Default recommendation: `base` for speed, `small` for quality.

## Commands

### Basic Transcription
```bash
whisper audio.mp3                          # Auto-detect language, base model
whisper audio.mp3 --model small            # Better accuracy
whisper audio.mp3 --language en            # Force English
whisper audio.mp3 --language zh            # Force Chinese
whisper audio.mp3 --language ja            # Force Japanese
```

### Output Formats
```bash
whisper audio.mp3 --output_format txt      # Plain text
whisper audio.mp3 --output_format srt      # Subtitles with timestamps
whisper audio.mp3 --output_format vtt      # WebVTT subtitles
whisper audio.mp3 --output_format json     # Full JSON with timestamps
whisper audio.mp3 --output_format all      # All formats at once
```

### Output Directory
```bash
whisper audio.mp3 --output_dir ~/transcripts/
```

### Transcribe from Video
```bash
whisper meeting.mp4 --model small --output_format txt
```

### Batch Transcription
```bash
whisper *.mp3 --model base --output_dir ./transcripts/
```

## Usage Examples

**"Transcribe meeting.m4a"**
```bash
whisper meeting.m4a --model small --output_format txt --output_dir ./
cat meeting.txt
```

**"Transcribe in Chinese"**
```bash
whisper recording.mp3 --language zh --model small
```

**"Generate subtitles for video.mp4"**
```bash
whisper video.mp4 --model medium --output_format srt --output_dir ./
```

**"Quick transcription (fast)"**
```bash
whisper audio.mp3 --model tiny
```

## OpenAI API Alternative
If local Whisper is too slow and you have an API key:
```bash
# Set key: export OPENAI_API_KEY=sk-...
curl https://api.openai.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -F file="@audio.mp3" \
  -F model="whisper-1"
```

## Rules
- Check if whisper is installed: `which whisper || python3 -m whisper --help 2>/dev/null`
- If not installed, show `pip install openai-whisper` and stop
- Check if ffmpeg is also installed: `which ffmpeg`
- Default to `base` model for files under 30 minutes, `small` for longer
- Warn user if using `medium` or `large` — processing may take several minutes
- Supported input formats: mp3, mp4, m4a, wav, ogg, flac, webm, mov
- Output transcript file path in the response
- For very long files (>1 hour), suggest splitting with ffmpeg first
- Never display raw JSON output — summarize and show the transcript text

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [claude-world](https://github.com/claude-world)
- **Source:** [claude-world/claude-agent](https://github.com/claude-world/claude-agent)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** yes
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-claude-world-claude-agent-speech-to-text
- Seller: https://agentstack.voostack.com/s/claude-world
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.