smart-transcriber

Transcribe meeting audio and generate structured notes using OpenAI APIs.

Install

pip install smart-transcriber

Quick Start

export OPENAI_API_KEY="sk-..."
transcribe meeting.mp3

This produces meeting.md with a summary, decisions, action items, speaker-labeled transcript, and more.

Requirements

• Python 3.11+
• OpenAI API key — set OPENAI_API_KEY environment variable
• ffmpeg — required for audio files larger than 25 MB (auto-detected)

Install ffmpeg: https://ffmpeg.org/download.html

Usage

transcribe <audio_file> [options]

Examples

# Basic transcription with summary
transcribe meeting.mp3

# Specify output path
transcribe meeting.mp3 --out notes.md --json-out notes.json

# Outline style instead of report
transcribe meeting.wav --style outline

# Use a different analysis model
transcribe meeting.m4a --analysis-model gpt-5-mini

# Transcription only (no summary/analysis)
transcribe meeting.mp3 --no-analysis

# Re-analyze a saved transcript
transcribe --analysis-only --transcript-input raw.json --out notes.md

# Render wall-clock timestamps
transcribe meeting.mp3 --start-time 09:30

# Provide hints for better results
transcribe meeting.mp3 --num-speakers 4 --prompt "Acme Corp, Project Phoenix"

Options

How It Works

The tool makes two OpenAI API calls per file:

1. Transcription — audio → text with timestamps (Whisper API)
2. Analysis — text → structured summary, speaker labels, decisions, action items (Chat API)

For large files (>25 MB), audio is automatically split into chunks using ffmpeg, transcribed separately, and merged.

Performance

Expect processing time to scale with audio duration. For faster results:

# Use a smaller analysis model
transcribe meeting.mp3 --analysis-model gpt-5-mini

# Skip analysis entirely
transcribe meeting.mp3 --no-analysis

License

MPL 2.0