Swahili Text-to-Speech API

A FastAPI-based REST API that provides text-to-speech (TTS) capabilities for Swahili text using fine-tuned Hugging Face models.

Features

• Converts Swahili text to speech using two different models:
  • Benny model: Benjamin-png/swahili-mms-tts-finetuned
  • Briget model: Benjamin-png/swahili-mms-tts-Briget_580_clips-finetuned
• Automatic Swahili language detection
• CORS support for cross-origin requests
• Efficient model caching
• Returns audio in WAV format

Prerequisites

pip install -r requirements.txt

Running the Server

To start the server:

python tts_linkedin.py

The server will run on http://0.0.0.0:8000

API Endpoints

1. Generate Speech using Benny Model

Endpoint: POST /tts/benny Request Body:

{
    "text": "Your Swahili text here"
}

2. Generate Speech using Briget Model

Endpoint: POST /tts/briget Request Body:

{
    "text": "Your Swahili text here"
}

Response

Both endpoints return:

• Content-Type: audio/wav
• Binary audio data in WAV format

Error Responses

• 400 Bad Request: If the provided text is not in Swahili
• Standard HTTP error codes for other failure cases

Example Usage

Using curl:

curl -X POST "http://localhost:8000/tts/briget" \
     -H "Content-Type: application/json" \
     -d '{"text":"Kijana huyu ni msataarabu sana sana"}' \
     --output output.wav

Using Python requests:

import requests

response = requests.post(
    "http://localhost:8000/tts/briget",
    json={"text": "Kijana huyu ni msataarabu sana sana"}
)

with open("output.wav", "wb") as f:
    f.write(response.content)

Technical Details

• Uses PyTorch for model inference
• Models are automatically loaded on first use and cached
• Runs on GPU if available, falls back to CPU
• Audio is generated at the model's native sampling rate
• Output is converted to 16-bit PCM WAV format

Implementation Notes

• The API uses lru_cache for efficient model loading
• Language detection is performed using the langdetect library
• CORS is configured to allow all origins, methods, and headers
• Audio processing uses scipy for WAV file generation
• The server uses FastAPI's automatic request validation with Pydantic models

Docker Support

You can also run the API using Docker:

# Build the Docker image
docker build -t swahili-tts .

# Run the container
docker run -d -p 8000:8000 swahili-tts