piper-dimits-dash

piper-dimits-dash is a web service for synthesizing audio from text using the Piper text-to-speech and Dimits library and streaming it using the DASH protocol. This service supports multiple languages and can dynamically generate audio files and DASH manifests.

Features

• Text-to-speech synthesis for multiple languages.
• Audio streaming using DASH protocol.
• Preload models on startup.
• Configurable model options through config.json.

Prerequisites

• Docker
• Python 3.x
• ffmpeg
• espeak-ng

Installation

1. Clone the repository:

git clone https://github.com/yourusername/piper-dimits-dash.git
cd piper-dimits-dash

2. Prepare config.json:

Create a config.json file in the root directory if you want to customize model options:

{
    "ru_RU": "ru_RU-denis-medium",
    "en_GB": "en_GB-alan-medium",
    "pl_PL": "pl_PL-darkman-medium"
}

3. Build the Docker image:

docker build -t piper-dimits-dash .

4. Run the Docker container:

docker run -p 8888:8888 -v $(pwd)/wav:/wav -v $(pwd)/models:/models piper-dimits-dash

To override the TEXT_MAX_LENGTH value:

docker-compose up -d --build --scale piper-dimits-dash=1 -e TEXT_MAX_LENGTH=500

Alternatively, you can use docker-compose:

docker-compose up

Usage

Endpoints

• POST /synthesise

  Synthesizes audio from text and returns the filename.

  Request body:

  {
      "language": "en_GB",
      "text": "Hello, world!",
      "lengthScale": 1.0,    // Optional
      "noiseScale": 0.3,     // Optional
      "noiseW": 1.0          // Optional
  }

  Response:

  {
      "status": "ok",
      "fileName": "generated_file_name"
  }

• POST /stream/dash

  Synthesizes audio and creates a DASH manifest for streaming.

  Request body:

  {
      "language": "en_GB",
      "text": "Hello, world!",
      "lengthScale": 1.0,    // Optional
      "noiseScale": 0.3,     // Optional
      "noiseW": 1.0          // Optional
  }

  Response:

  {
      "status": "ok",
      "manifestName": "generated_manifest_name.mpd"
  }

• GET /get_stream/<file_name>

  Retrieves the DASH manifest or audio segments.

  Response:

  Returns the requested file.

Request Validation

The POST /synthesise and POST /stream/dash endpoints require a request body with specific fields. These fields are validated for type and value ranges using Pydantic models.

Request Body Fields:

• language: A string representing the language code (e.g., “en_GB”). Maximum length is 20 characters.

• text: The text to be synthesized. Maximum length is defined by the TEXT_MAX_LENGTH environment variable (default is 255 characters).

• lengthScale: An optional float value representing the length scale. Default is 1.0, with a valid range between 0.1 and 5.0.

• noiseScale: An optional float value representing the noise scale. Default is 0.3, with a valid range between 0.0 and 1.0.

-noiseW: An optional float value representing the noise weight. Default is 1.0, with a valid range between 0.0 and 1.0.

Configuration

Model options are defined in config.json. If this file does not exist, default options will be used:

{
    "en_GB": "en_GB-alan-medium",
}

Code Overview

• load_model_options: Loads model configurations from config.json.
• preload_models: Preloads TTS models on startup.
• get_hash_name: Generates a SHA-1 hash for the given text.
• synthesise_audio_to_file: Synthesizes audio and saves it to a file.
• synthesis: Endpoint for synthesizing audio from text.
• stream_dash: Endpoint for creating DASH manifest and streaming audio.
• get_stream: Serves the DASH manifest and audio segments.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgements

• Piper
• Dimits
• Bottle Web Framework
• FFmpeg

Contributing

Feel free to submit issues, fork the repository and send pull requests. For major changes, please open an issue first to discuss what you would like to change.

---

Enjoy using piper-dimits-dash! If you have any questions or suggestions, please let us know.