bark_kubernetes

README

Bark on Kubernetes with wfl

This project demonstrates how to harness the power of Bark to generate engaging audio from text prompts on Kubernetes (here GKE). Bark is a transformer-based text-to-audio model from Suno that can generate realistic speech, background noise, music, and even nonverbal communications like laughter and sighs.

This example requires a GKE cluster with a V100 GPU. Feel free to test it on AKS or EKS by setting "distribution" in the JobTemplate accordingly.

Project Structure

• bark.go: The main Go application that uses wfl and Bark on Kubernetes.
• Dockerfile: Used to create a container with Bark.
• Makefile: Helps build, push, and run the Bark container.
• cloudbuild.yaml: Configuration file for Google Cloud Build.
• bark.py: Embedded Bark script for text-to-audio conversion.

Getting Started

1. Build the Docker image:

   make build
   

2. Push the Docker image to Google Container Registry:

   export GOOGLE_PROJECT=your-google-project-id
   make push
   

3. Run the Go application:

   Set the other environment variables (see bark.go).

   go run bark.go
   

4. Check the generated audio file at https://raw.githubusercontent.com/dgruber/wfl/master/examples/bark_kubernetes/cow.wav.

Resources

• wfl GitHub Repository
• Bark GitHub Repository

Support and Troubleshooting

For any issues or questions related to wfl, please refer to the wfl GitHub repository for support and troubleshooting.