AI4Bharat Indic-TTS

repository·master·Indexed 18 days ago

A text-to-speech system providing state-of-the-art models for 15 major Indian languages, including Assamese, Bengali, Hindi, Tamil, and Telugu. It utilizes a unified architecture based on FastPitch acoustic models and HiFi-GAN V1 vocoders. The repository includes tools for local inference, hosting a REST API server via uvicorn, deploying to Azure Machine Learning using Triton server, and training models from datasets formatted in LJSpeech style.

Tokens: 11.2K
Snippets: 40
Records: 49
Agent score: 64%

What's inside Indic-TTS

Organize language checkpoints for deployment
master
To optimize GPU RAM usage, models are grouped by language families. Place your language-specific checkpoint folders in the corresponding directory structure before deployment:
- North-Indian (Indo-Aryan) languages (as, bn, gu, hi, mr, or, pa, raj): Place in inference/checkpoints/indo-aryan/checkpoints
- South-Indian (Dravidian) languages (kn, ml, ta, te): Place in inference/checkpoints/dravidian/checkpoints
- Miscellaneous languages (en, brx, mni): Place in inference/checkpoints/misc/checkpoints
https://github.com/ai4bharat/indic-tts/blob/master/inference/triton_server/azure_ml/README.md
Set up the Azure Machine Learning (AML) environment
master
Before deploying, navigate to the Triton server directory and set the required environment variables for your Azure resource group, workspace, and Docker registry. You must also ensure that any relevant .yml configuration files are updated to match these values.
```
cd inference/triton_server

export RESOURCE_GROUP=Dhruva-prod
export WORKSPACE_NAME=dhruva--central-india
export DOCKER_REGISTRY=dhruvaprod
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/triton_server/azure_ml/README.md
Run the Triton server container
master
Start the Triton server using the tts_triton image. Ensure you provide sufficient shared memory (--shm-size=256m), enable GPU access (--gpus=1), and mount your local checkpoints/ directory to /models/checkpoints inside the container so the server can access the models. The server exposes port 8000.
```
docker run --shm-size=256m --gpus=1 --rm -v ${PWD}/checkpoints/:/models/checkpoints -p 8000:8000 -t tts_triton
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/triton_server/README.md
Create the Azure ML execution environment
master
Use the Azure CLI to create an execution environment based on the provided environment.yml file.
```
az ml environment create -f azure_ml/environment.yml -g $RESOURCE_GROUP -w $WORKSPACE_NAME
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/triton_server/azure_ml/README.md
Push the Triton Docker image to Azure Container Registry
master
To make the Triton server image available for deployment, tag the local tts_triton image with your registry's URI and push it to the Azure Container Registry (ACR).
```
az acr login --name $DOCKER_REGISTRY
docker tag tts_triton $DOCKER_REGISTRY.azurecr.io/tts/triton-tts-coqui:latest
docker push $DOCKER_REGISTRY.azurecr.io/tts/triton-tts-coqui:latest
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/triton_server/azure_ml/README.md
Test online inference with client.py
master
To verify the deployment:
1. Retrieve the endpoint domain (exclude https:// and trailing /) and the authentication key from the 'Consume' tab in Azure ML Studio.
2. Configure client.py with the following settings:
  Set ENABLE_SSL = True
  Set ENDPOINT_URL to your domain.
  Set the Authorization value inside HTTP_HEADERS using your key.
3. Execute the client script.
```
# client.py configuration steps
ENABLE_SSL = True
ENDPOINT_URL = "your-endpoint-domain"
HTTP_HEADERS = {"Authorization": "your-auth-key"}

# Run via terminal:
# python3 client.py
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/triton_server/azure_ml/README.md
Set up AI4Bharat-TTS Inference environment
master
To use the AI4Bharat-TTS inference models, ensure you have Python 3.9+ installed. Follow these steps to prepare your environment:
1. Install System Dependencies (Linux only): Navigate to the inference directory and install the required system libraries using apt-get.
2. Install Python Dependencies: Install the necessary machine learning and utility packages using pip.
3. Download and Prepare Models: Download the model checkpoints from the official releases. Create a directory named checkpoints in your project root and unzip the downloaded files there.
```
# 1. System dependencies (Linux)
cd inference
sudo apt-get install libsndfile1-dev ffmpeg enchant

# 2. Python dependencies
pip install -r requirements-ml.txt requirements-utils.txt

# 3. Model setup
mkdir checkpoints
# [Download and unzip models into the checkpoints/ folder]
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/README.md
Build the Triton server Docker image
master
To build the Docker image for the Triton server, navigate to the inference/ directory and use the Dockerfile located in triton_server/. Tag the resulting image as tts_triton.
```
cd inference/
docker build -f triton_server/Dockerfile -t tts_triton .
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/triton_server/README.md
Install Indic-TTS via Environment Setup
master
To set up the Indic-TTS environment, follow these steps to install system dependencies, create a Conda environment, and install the required Python packages (PyTorch, Trainer, and TTS).
1. System Dependencies
Install the necessary system libraries using apt-get:
```
sudo apt-get install libsndfile1-dev ffmpeg enchant
```
2. Conda Environment
Create and activate a new Conda environment named tts-env:
```
conda create -n tts-env
conda activate tts-env
```
3. Install PyTorch
Install PyTorch with CUDA 11.3 support:
```
pip3 install -U torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
```
4. Install Trainer
Clone the Trainer repository and install it in editable mode with all dependencies:
```
git clone https://github.com/gokulkarthik/Trainer 
cd Trainer
pip3 install -e .[all]
cd ..
```
Note: If you are manually copying files instead of cloning, ensure you apply the specific fixes mentioned in the repository documentation for wandb_logger.py, trainer.py, and distribute.py.
5. Install TTS
Clone the TTS repository and install it in editable mode with all dependencies:
```
git clone https://github.com/gokulkarthik/TTS 
cd TTS
pip3 install -e .[all]
cd ..
```
6. Final Requirements
Install any remaining requirements from the project's requirements file:
```
pip3 install -r requirements.txt
```
```
# 1. Create environment
sudo apt-get install libsndfile1-dev ffmpeg enchant
conda create -n tts-env
conda activate tts-env

# 2. Setup PyTorch
pip3 install -U torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113

# 3. Setup Trainer
git clone https://github.com/gokulkarthik/Trainer 
cd Trainer
pip3 install -e .[all]
cd ..

# 4. Setup TTS
git clone https://github.com/gokulkarthik/TTS 
cd TTS
pip3 install -e .[all]
cd ..

# 5. Install other requirements
pip3 install -r requirements.txt
```
https://github.com/ai4bharat/indic-tts/blob/master/README.md
Host a REST API server for TTS
master
You can host the TTS models as a REST API service using uvicorn. First, install the server-specific requirements, then start the API server.
```
pip install -r requirements-server.txt
uvicorn server:api
```
https://github.com/ai4bharat/indic-tts/blob/master/inference/README.md
Run TTS inference locally
master
Once the environment is set up and checkpoints are placed in the checkpoints folder, you can run inference. For a concrete implementation example, refer to the sample.py file in the repository.
https://github.com/ai4bharat/indic-tts/blob/master/inference/README.md
Prepare IndicTTS data for training
master
Before training, you must prepare your dataset using the provided preprocessing notebooks:
1. Format Dataset: Convert the IndicTTS dataset into the LJSpeech format using preprocessing/FormatDatasets.ipynb.
2. Analyze Dataset: Verify the suitability of the dataset for TTS using preprocessing/AnalyzeDataset.ipynb.
https://github.com/ai4bharat/indic-tts/blob/master/README.md

AI4Bharat Indic-TTS

What's inside Indic-TTS

1. System Dependencies

2. Conda Environment

3. Install PyTorch

4. Install Trainer

5. Install TTS

6. Final Requirements