transcriptionEvergreen

Pipeline that uses whisperAI to transcribe the focus group meetings

The bash script does:

1. Covert audio to 16 kHz mono WAV
2. Runs whisper.cpp to generate plain text scripts
3. Runs Pyannote-Whisper to add speaker labels to each segment

You end up with, for every audio/<name>.WAV:

audio/<name>.txt

audio/<name>_diarized.txt

Instructions for use

1. Clone the repo

#!/bin/bash
git clone https://github.com/your-org/whisper-transcription.git
cd whisper-transcription

2. Build the whisper.cpp

#!/bin/bash
cd whisper.cpp
make
./models/download-ggml-model.sh large-v1
cd ..

3. Install the python dependencies

#!/bin/bash
python3 -m venv venv
source venv/bin/activate
pip install pywhispercpp pyannote-audio pyannote-whisper

Usage

1. Add the audio files into audio directory (wav/WAV files through git cp)
2. Run the pipeline

#!/bin/bash
chmod +x run_pipeline.sh
./run_pipeline.sh en

Output

Now under audio directory each WAV file should have a corresponding txt file