Pipescribe

Pipescribe is a real-time audio transcription tool that captures audio from PipeWire sources and transcribes it using the Whisper speech recognition model.

Features

• Capture audio from any PipeWire source (source detection or manual selection via regex patterns)
• Real-time speech-to-text transcription
• Support for multiple languages

Requirements

• Rust (cargo)
• PipeWire
• Whisper model files

Installation

1. Clone the repository:

git clone https://github.com/mz2/pipescribe.git
cd pipescribe

2. Download a Whisper model file:

mkdir -p models
./bin/download-ggml-models.sh medium.en

(model download script whisper.cpp(lifted from https://github.com/ggml-org/whisper.cpp/blob/master/models/download-ggml-model.sh)

3. Build the project:

cargo build --release

Usage

Basic usage:

pipescribe --buffer-seconds 5 --model ./models/ggml-medium.en.bin

... or to test out a local build:

cargo run --bin pipescribe -- --buffer-seconds 5 --model ./models/ggml-medium.en.bin