Pipecat Deepdub TTS

Official Deepdub AI Text-to-Speech integration for Pipecat -- a framework for building voice and multimodal conversational AI applications.

Note: This integration is maintained by Deepdub AI. As the official provider of the TTS service, we are committed to actively maintaining and updating this integration.

Pipecat Compatibility

Tested with Pipecat v0.0.97

This integration has been tested with Pipecat version 0.0.97. For compatibility with other versions, refer to the Pipecat changelog.

Features

• Real-time Streaming: WebSocket-based streaming for low-latency audio generation
• Multiple Models: Support for Deepdub TTS models (dd-etts-2.5, dd-etts-3.0)
• Voice Customization: Configurable temperature, variance, tempo, and prompt boost
• Accent Control: Blend accents between locales with fine-grained ratio control
• Flexible Sample Rates: 8000, 16000, 22050, 24000, 44100, 48000 Hz
• Interruption Handling: Clean disconnect/reconnect on user interruption
• Metrics Support: Built-in performance tracking and monitoring

Installation

Using pip

pip install pipecat-deepdub-tts

Using uv

uv add pipecat-deepdub-tts

From source

git clone https://github.com/deepdub-ai/pipecat-deepdub-tts.git
cd pipecat-deepdub-tts
pip install -e .

Quick Start

1. Get Your Deepdub API Key

Sign up at Deepdub AI and obtain your API key.

2. Basic Usage

from pipecat_deepdub_tts import DeepdubTTSService

tts = DeepdubTTSService(
    api_key="your-deepdub-api-key",
    voice_id="your-voice-prompt-id",
    model="dd-etts-2.5",
)

3. Complete Example with Pipeline

import asyncio
import os
from dotenv import load_dotenv
from pipecat.pipeline.pipeline import Pipeline
from pipecat.pipeline.runner import PipelineRunner
from pipecat.pipeline.task import PipelineTask, PipelineParams
from pipecat.services.openai.llm import OpenAILLMService
from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
from pipecat_deepdub_tts import DeepdubTTSService

load_dotenv()

async def main():
    tts = DeepdubTTSService(
        api_key=os.getenv("DEEPDUB_API_KEY"),
        voice_id=os.getenv("DEEPDUB_VOICE_ID"),
        model=os.getenv("DEEPDUB_MODEL", "dd-etts-2.5"),
    )

    llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"))

    messages = [
        {"role": "system", "content": "You are a helpful assistant."},
    ]
    context = OpenAILLMContext(messages)
    context_aggregator = llm.create_context_aggregator(context)

    pipeline = Pipeline([
        llm,
        tts,
        context_aggregator.assistant(),
    ])

    task = PipelineTask(pipeline, params=PipelineParams(enable_metrics=True))
    runner = PipelineRunner()
    await runner.run(task)

if __name__ == "__main__":
    asyncio.run(main())

Configuration

DeepdubTTSService Constructor

Parameter	Type	Default	Description
api_key	str	required	Deepdub API key for authentication
voice_id	str	required	Voice prompt ID for TTS synthesis
model	str	"dd-etts-2.5"	TTS model name
sample_rate	int	16000	Audio sample rate in Hz
params	InputParams	None	Optional voice customization parameters

InputParams

Parameter	Type	Default	Description
locale	str	"en-US"	Language locale for synthesis
temperature	float	None	Controls output variability
variance	float	None	Controls variance in generated speech
tempo	float	None	Speech tempo multiplier
prompt_boost	bool	None	Enable prompt boosting for improved quality
accent_base_locale	str	None	Base locale for accent control
accent_locale	str	None	Target accent locale
accent_ratio	float	None	Accent blending ratio (0.0 to 1.0)

Example with Custom Configuration

tts = DeepdubTTSService(
    api_key="your-api-key",
    voice_id="your-voice-prompt-id",
    model="dd-etts-3.0",
    sample_rate=24000,
    params=DeepdubTTSService.InputParams(
        locale="en-US",
        temperature=0.7,
        tempo=1.1,
        prompt_boost=True,
        accent_base_locale="en-US",
        accent_locale="fr-FR",
        accent_ratio=0.3,
    ),
)

Environment Variables

Create a .env file in your project root:

# Deepdub TTS
DEEPDUB_API_KEY=your_deepdub_api_key_here
DEEPDUB_VOICE_ID=your_voice_prompt_id_here
DEEPDUB_MODEL=dd-etts-2.5

# OpenAI (for LLM in examples)
OPENAI_API_KEY=your_openai_api_key_here

# AssemblyAI (for STT in examples)
ASSEMBLYAI_API_KEY=your_assemblyai_api_key_here

Examples

See the examples directory for complete working examples:

• deepdub_tts_basic.py -- Full pipeline with AssemblyAI STT, OpenAI LLM, and Deepdub TTS

To run the example:

# Install dependencies
pip install pipecat-deepdub-tts "pipecat-ai[assemblyai,openai,silero]" python-dotenv

# Set up your .env file with API keys (see .env.example)
# Then run
python examples/foundational/deepdub_tts_basic.py

Testing

Unit tests (no API key needed)

pytest tests/test_tts.py -k "not Integration"

Integration tests (saves audio to disk)

Requires DEEPDUB_API_KEY and DEEPDUB_VOICE_ID environment variables:

# Load .env and run integration tests
pytest tests/test_tts.py -k "Integration" -s

Generated audio files are saved to tests/output/ for manual listening verification.

Requirements

• Python >= 3.10, < 3.13
• deepdub >= 0.1.20
• pipecat-ai >= 0.0.97, < 0.1.0
• websockets >= 15.0.1, < 16.0
• loguru >= 0.7.3

Architecture

DeepdubTTSService extends Pipecat's InterruptibleTTSService base class for WebSocket-based TTS without word-level timestamp alignment. It uses the DeepdubClient from the deepdub package for WebSocket protocol communication.

The service:

1. Opens a WebSocket to Deepdub's streaming endpoint on pipeline start
2. Sends text chunks as they arrive from the LLM via async_stream_text()
3. Receives raw PCM audio (s16le format) via a background task
4. Pushes TTSAudioRawFrame frames into the pipeline
5. Handles interruptions by disconnecting and reconnecting

License

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

Support

• GitHub Issues: pipecat-deepdub-tts/issues
• Deepdub: deepdub.ai
• Pipecat Discord: discord.gg/pipecat