neowisard/wyoming-faster-whisper

Wyoming protocol for Home Assistant with Systran original faster whisper and OpenAI LARGE V3 models

Wyoming Faster Whisper for Large v3 model with official faster whisper on CUDA

Wyoming protocol server for the faster-whisper speech to text system.

Pre-requisites :

NVIDIA GPU, driver , cuda installed

Home Assistant Add-on

Source

Local Install

Set up Python virtual environment:

Virtual Python env (Only for Conda using)

conda create -n fwf python=3.10
conda activate fwf

Clone the repository

git clone https://github.com/neowisard/wyoming-faster-whisper.git
cd wyoming-faster-whisper
pip install -r requirements.txt

Download large model to data (/ai/models/whisper) dir

git clone https://huggingface.co/Systran/faster-whisper-large-v3
mv fwhisper-large-v3 large-v3

or quantized to INT8 (smaller and faster Largev3)

git clone https://huggingface.co/neowisard/fwhisper-large-v3-int8
mv fwhisper-large-v3-int8 large-v3i

Run a server anyone can connect to: Full precision

python3 -m wyoming_faster_whisper --uri 'tcp://0.0.0.0:10300' --data-dir /ai/models/whisper --model large-v3 --beam-size 1 --language ru --download-dir /ai/models/whisper --compute-type float16 --device cuda --initial-prompt "promt"

Quantized INT8

python3 -m wyoming_faster_whisper --uri 'tcp://0.0.0.0:10300' --data-dir /ai/models/whisper --model large-v3i --beam-size 1 --language ru --download-dir /ai/models/whisper --compute-type int8_float32 --device cuda

Source