format37 / vosk-api-gpu

Vosk ASR Docker images with GPU for Jetson Nano / Xavier NX and PCs

Vosk API - Docker/GPU

Docker images with GPU for Jetson Nano / Xavier NX boards and PCs with NVIDIA cards.

Usage

Pull an existing image with a required tag.

docker pull sskorol/vosk-api:$TAG

Use it as a baseline in your app's Dockerfile:

FROM sskorol/vosk-api:$TAG

Building

Clone sources and check a build file help:

git clone https://github.com/format37/vosk-api-gpu.git
cd vosk-api-gpu
./build.sh -h

Then run it with required args depending on your platform, e.g.:

./build.sh -m nano -i ml -t 0.3.27

Pc (check latest version at https://hub.docker.com/r/sskorol/vosk-api/tags?page=1&ordering=last_updated):

docker pull sskorol/vosk-api:0.3.27-pc

You can check the available NVIDIA base image tags here and here.

update ip to your host's ip in docker-compose.yml

ports:
      - 10.2.5.212:2700:2700

Note that you have to download and extract a required model into ./model folder.

wget https://alphacephei.com/vosk/models/vosk-model-ru-0.10.zip
unzip vosk-model-ru-0.10.zip
mv vosk-model-ru-0.10 model

To build images for PC, use the following script:

./build-pc.sh

It'll build 2 images: base and a sample Vosk server.

To start a newly built container, run the following command:

docker-compose up -d

Also make sure you have at least Docker (20.10.6) and Compose (1.29.1) versions.

Your host's CUDA version should match the container's. Jetson images were built with CUDA 10.1. As per the desktop version: CUDA 11.3.0 was used.