SoundScribe

Audio transcription CLI tool using faster-whisper with a progress bar.

I got tired of running faster-whisper and staring at a blank terminal with no idea whether it was working, stuck, or halfway done. Long audio files could take minutes with zero feedback. So I wrapped it in a proper CLI with a live progress bar and time estimates — because knowing how long something will take is basic UX, even for a local dev tool.

Plus, I wanted something that does not reach out to the cloud to do the transcription.

Features

• Fast transcription using faster-whisper
• Progress bar with time estimates
• Supports various audio formats (.m4a, .mp3, .wav, etc.)
• Automatic output to .txt file

Supported Audio Formats

soundscribe relies on faster-whisper, which uses ffmpeg for audio decoding. Any format ffmpeg supports will work. Common formats include:

Format	Extension
MPEG Audio Layer 3	.mp3
MPEG-4 Audio	.m4a
Waveform Audio	.wav
FLAC	.flac
Ogg Vorbis	.ogg
Opus	.opus
WebM	.webm
MP4 video (audio extracted)	.mp4
AAC	.aac

Note: ffmpeg must be installed and available on your PATH for formats other than WAV.

Installation

Install with pip (editable)

cd soundscribe
pip install -e .

First Run

On first run, faster-whisper will download the model from Hugging Face and cache it locally. You'll see download progress and a warning like:

Warning: You are sending unauthenticated requests to the HF Hub...

This is normal. The model is only downloaded once — subsequent runs load it from the local cache (~/.cache/huggingface/) with no network activity. Your audio is never sent anywhere.

To suppress the warning and get faster downloads, set a (free) Hugging Face token:

export HF_TOKEN=hf_your_token_here

Usage

Once installed, you can use the soundscribe command from anywhere:

soundscribe audio_file.m4a

This will:

1. Transcribe the audio file
2. Show a progress bar during transcription
3. Save the transcript as audio_file.txt in the same directory

Example

$ soundscribe my_recording.m4a
✨ Transcribing... ████████████████████ 100% 0:00:45 0:00:00
✨ Transcript saved to: my_recording.txt

License

MIT — see LICENSE for details.