Local audio transcription with OpenAI Whisper. No API keys, no cloud, no data leaving your machine.

You have an audio file you want transcribed. The usual options:

• OpenAI / AssemblyAI / Deepgram APIs - fast, accurate, but your audio goes to a server.
• macOS dictation / Otter.ai - also cloud.
• This - Whisper running entirely on your machine. Sensitive audio (interviews, medical notes, legal recordings) never leaves.

• Python 3.7+
• FFmpeg (for audio decoding)
• ~1-10 GB disk space depending on which Whisper model you load

pip install openai-whisper ffmpeg-python torchaudio

Install FFmpeg if you don't have it:

# macOS
brew install ffmpeg

# Ubuntu / Debian
sudo apt install ffmpeg

# Windows (with Chocolatey)
choco install ffmpeg

Verify:

ffmpeg -version

Edit text-to-speech.py to point at your audio file:

audio_path = "/path/to/your/audio.wav"

Then run:

python text-to-speech.py

Output: transcript printed to terminal and saved as a .txt file alongside the audio.

Change the model in the script:

model = whisper.load_model("small")  # or "base", "medium", "large"

The first time you load a model, Whisper downloads it from the OpenAI CDN.

Anything FFmpeg can decode: WAV, MP3, M4A, FLAC, OGG, AAC, WMA, etc.

FFmpeg errors - verify the install with ffmpeg -version. On Windows, make sure C:\ffmpeg\bin is in your PATH.

FileNotFoundError - use an absolute path to the audio file. Check for typos in the filename.

Model download fails - first run requires internet to fetch the model. Check disk space too (models range from ~150 MB to ~3 GB).

Poor accuracy - try a larger model. Accuracy on noisy or accented audio scales heavily with model size. The large model handles non-English audio much better than base.

MIT