Pipecat integration

Learn how to integrate Speechmatics STT with Pipecat.

Pipecat is an open-source framework for building voice agents. When Speechmatics STT is integrated with Pipecat, you can build real-time voice and multimodal conversational agent specifically tailored to your needs.

Pipecat is perfect for:

• Voice AI: Voice assistants, chatbots, and IVR systems
• Transcription: Realtime transcription of live events or media
• Accessibility applications: Screen readers and assistive technologies
• Content creation: Podcasts, dubbing, audiobooks, and voice-overs
• Media production: News broadcasts and automated announcements

Features

• Realtime transcription: low-latency speech-to-text for responsive agents
• Speaker diarization: track who’s speaking in multi-participant sessions
• Turn detection: capture natural speech boundaries automatically
• Noise robustness: maintain accuracy in challenging environments
• Custom vocabularies: boost recognition for domain-specific terms
• Flexible deployment: use on-device, cloud, or hybrid Pipecat setups

Quickstart

Requirements

• Python 3.10 or later
• uv package manager installed
• Pipecat >= 1.2
• Speechmatics account. You can create one here.
• Speechmatics API key. You can generate one in the Portal.

Installation

pip install "pipecat-ai[speechmatics]"

Usage

Set the environment variable SPEECHMATICS_API_KEY to your Speechmatics API key.

export SPEECHMATICS_API_KEY=your_api_key

import asyncio
import os

from pipecat.services.speechmatics import SpeechmaticsSTTService


async def main():
    stt = SpeechmaticsSTTService(
        api_key=os.environ["SPEECHMATICS_API_KEY"],
    )

    async def audio_stream():
        # Replace with your real audio source.
        yield from [b"fake_audio_chunk_1", b"fake_audio_chunk_2"]

    async for result in stt.transcribe(audio_stream()):
        speaker = f"Speaker {result.speaker}" if result.speaker else "Unknown"
        print(f"{speaker}: {result.text}")


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

For detailed examples, please see the Speechmatics Academy.