Flask Speech-to-Text App with Whisperx

This repository contains a Flask application that uses the Whisperx library to transcribe and align audio data, and then assign speaker labels to the transcriptions.

Setup

1. Build Docker Image

To build the Docker image, navigate to the directory containing the Dockerfile and run:

   docker build -t whisperx-app:latest .

2. Run Docker Container

After building the Docker image, run the container using:

   docker run -p 5000:5000 -d whisperx-app:latest

Usage

To use the app, send a POST request to http://localhost:5000/transcribe with an audio file attached with the key audio. The application will return the transcriptions with speaker labels.

Example:

curl -X POST -F 'audio=@your_audio_file.wav' http://localhost:5000/transcribe

Requirements

• Docker
• Python 3.10
• Whisperx

Environment Variables

• CUDA_VISIBLE_DEVICES: This environment variable is used to limit the number of visible CUDA devices. In this application, it's set to an empty string which implies no GPU is used.
• NVIDIA_VISIBLE_DEVICES: Similar to CUDA_VISIBLE_DEVICES, this environment variable is used to limit the number of visible Nvidia devices. In this application, it's set to an empty string.
• HUGGING_FACE_TOKEN: The API token for the Hugging Face API. Set this to your Hugging Face API token.

Hosting on Kubernetes

If you prefer to host the application on a Kubernetes cluster, configuration files are provided in the k8-manifests directory.

1. Deploy the Application

   To deploy the application, use the kubectl apply command:

   kubectl apply -f k8-manifests/deployment.yaml
   kubectl apply -f k8-manifests/service.yaml
   kubectl apply -f k8-manifests/hpa.yaml` 

2. Access the Application

   The application will be accessible through the service whisperx-transcription-service at port 80.

Notes

• The application runs on the CPU and not the GPU, so it may not be suitable for large-scale production use.
• Remember to replace <YOUR-TOKEN> in the Dockerfile with your actual Hugging Face API token.