A powerful and easy-to-use tool for transcribing podcast MP3 files into clean, readable transcripts using OpenAI's Whisper AI model.

• High Accuracy: Uses OpenAI's Whisper model for state-of-the-art transcription
• Batch Processing: Transcribe entire folders of MP3 files at once
• Multiple Formats: Save transcripts as TXT, JSON, and SRT (subtitle) formats
• Timestamps: Optional timestamps for each segment
• Language Support: Auto-detects language or specify manually
• Clean Output: Beautiful, readable transcripts
• Progress Tracking: Visual progress bars for long transcriptions
• Flexible Models: Choose from 5 model sizes (speed vs accuracy)

• Python 3.8 or higher
• macOS, Linux, or Windows
• At least 2GB RAM (more for larger models)
• Internet connection (first time only, to download models)

1. Clone or download this repository

2. Install dependencies:

   pip install -r requirements.txt

   Note for macOS with Apple Silicon (M1/M2/M3): If you have issues, install PyTorch first:

   pip install torch torchaudio
   pip install -r requirements.txt

3. Verify installation:

   python transcribe.py --help

Transcribe all MP3 files in a folder:

python transcribe.py /path/to/your/podcasts

This will:

• Find all MP3 files in the specified folder
• Transcribe each file using the base model
• Save transcripts to a transcripts/ folder
• Generate both TXT and JSON formats

Use a more accurate model:

python transcribe.py /path/to/podcasts --model medium

Specify output directory:

python transcribe.py /path/to/podcasts --output my_transcripts

Specify language (faster than auto-detect):

python transcribe.py /path/to/podcasts --language en

Choose output formats:

# Save as TXT only
python transcribe.py /path/to/podcasts --formats txt

# Save all formats (TXT, JSON, SRT)
python transcribe.py /path/to/podcasts --formats txt json srt

Complete example:

python transcribe.py ~/Downloads/podcasts \
  --model medium \
  --output transcripts \
  --language en \
  --formats txt json srt

Choose the right model for your needs:

Recommendation: Start with base model. Upgrade to medium or large if you need better accuracy.

Clean, readable text with two sections:

1. Transcript with timestamps for each segment
2. Clean transcript without timestamps (perfect for reading)

Example:

Transcript: podcast_episode_123.mp3
Generated: 2025-11-04 10:30:00
================================================================================

[00:00 -> 00:05] Welcome to the show, today we're talking about AI.
[00:05 -> 00:12] This is a fascinating topic that affects everyone.

================================================================================
CLEAN TRANSCRIPT (no timestamps):

Welcome to the show, today we're talking about AI. This is a fascinating topic that affects everyone.

Structured data with full details including:

• Audio filename
• Detected language
• Complete text
• All segments with timestamps
• Generation timestamp

Standard subtitle format that can be used with video players or video editing software.

Whisper supports 99+ languages. Common language codes:

• en - English
• es - Spanish
• fr - French
• de - German
• it - Italian
• pt - Portuguese
• ja - Japanese
• ko - Korean
• zh - Chinese

Or leave blank for auto-detection!

1. First Run: The first time you run the tool, it will download the selected Whisper model (~100MB-3GB depending on size). This happens once per model.

2. Audio Quality: Higher quality audio = better transcriptions. Whisper is very robust though!

3. Processing Time:

   • A 1-hour podcast takes ~5-10 minutes with base model
   • Use tiny for quick tests
   • Use medium or large for important content

4. Batch Processing: Process multiple files overnight if you have many podcasts.

5. Storage: Transcripts are small (typically <100KB per hour of audio).

pip install openai-whisper

Use a smaller model:

python transcribe.py /path/to/podcasts --model tiny

• Use a smaller model (tiny or base)
• Ensure no other heavy processes are running
• Check if your CPU supports optimizations

• Try a larger model (medium or large)
• Ensure audio quality is good
• Specify the language explicitly

# 1. Put your podcast MP3s in a folder
mkdir ~/podcasts_to_transcribe
# (copy your MP3 files there)

# 2. Run the transcription
python transcribe.py ~/podcasts_to_transcribe --model base

# 3. Find your transcripts
ls transcripts/

# 4. Open and enjoy!
open transcripts/podcast_episode.txt

Found a bug or want a feature? Feel free to:

• Open an issue
• Submit a pull request
• Share your feedback

MIT License - feel free to use this tool for personal or commercial projects!

• Built with OpenAI Whisper
• Inspired by the need for accessible podcast transcripts

If you find this tool useful, please star the repo and share it with others!

Questions? Open an issue or check the Whisper documentation.