livekit-plugins-munsit

LiveKit Agents plugin for Munsit voice AI: Arabic speech-to-text (batch and live streaming) and Arabic text-to-speech.

This package consolidates the former livekit-plugins-faseeh (TTS) — that name now ships a deprecation shim that installs this package.

Installation

pip install livekit-plugins-munsit

Set your API key (from the Munsit dashboard):

export MUNSIT_API_KEY=your-key

Speech-to-Text

from livekit.agents import AgentSession
from livekit.plugins import munsit, silero

session = AgentSession(
    stt=munsit.STT(mode="streaming"),   # live transcription over WS /api/v1/listen
    vad=silero.VAD.load(),
)

Two modes:

Streaming options:

munsit.STT(
    mode="streaming",
    endpointing_ms=800,          # silence that ends a turn (100–5000; retunable live)
    smart_turn=True,             # semantic end-of-turn gating
    hotwords=["فرحان", "تطبيق منصت"],   # custom vocabulary (≤200 entries, ≤40 chars)
    interim_results=True,
    correlation_id="call-1234",
    on_enrichment=lambda ev: print(ev),  # per-turn Sentiment / Gender events
)

Batch extras: hotwords=[...] and return_confidence=True (per-word confidence on timestamps). hotwords only bias the Arabic munsit model — with munsit-en-ar (the default) they are ignored with a warning, in both modes.

Text-to-Speech

from livekit.plugins import munsit

tts = munsit.TTS(
    voice_id="ar-hijazi-female-1",
    model="faseeh-v1-preview",
)

MUNSIT_API_KEY is used for both engines (the legacy FASEEH_API_KEY env var still works for TTS).

Documentation

Full API reference: https://docs.munsit.com — see Speech-to-Text Streaming for the /api/v1/listen protocol (events, close codes, billing).

Development

uv sync
uv run pytest tests/ -k "not Integration"   # offline suite (fake server)
MUNSIT_API_KEY=... uv run pytest tests/     # includes live integration tests

License

Apache-2.0