Palabra plugin for LiveKit Agents

Support for Palabra realtime Speech-to-Text and Text-to-Speech APIs, using WebSocket streaming interfaces.

See STT documentation and TTS documentation for more information.

Installation

pip install livekit-plugins-palabra

Pre-requisites

The Palabra plugin requires an API key to authenticate. You can create one at platform.palabra.ai/api-keys.

Set the API key in your .env file:

PALABRA_API_KEY=<your_palabra_api_key>

Usage

Speech-to-Text (STT)

Use Palabra STT in an AgentSession or as a standalone streaming transcription service. The spoken language is auto-detected by default:

from livekit.agents import AgentSession
from livekit.plugins import palabra

session = AgentSession(
    stt=palabra.STT(),
    # ... llm, tts, etc.
)

Live translation

Palabra STT can translate speech on the fly. With translate_languages set, each emitted SpeechData maps as: language/text = the translation; source_languages/source_texts = the original speech. Final transcripts arrive already in the target language, so a translating agent works without an LLM in the loop:

stt = palabra.STT(translate_languages=["es"])

Text-to-Speech (TTS)

Use Palabra TTS for real-time speech synthesis:

from livekit.plugins import palabra

session = AgentSession(
    tts=palabra.TTS(
        language="en",
        voice_id="default_low",  # or "default_high", or any Palabra voice id
        speed=0.5,  # 0.0-1.0; 0.5 is natural conversational pace
    ),
    # ... stt, llm, etc.
)

The TTS streams over one persistent WebSocket session: text chunks coming from the LLM are tokenized into sentences and synthesized as they arrive, and an interruption cancels the in-flight synthesis server-side.