Sign inContact usTry Munsit freeENAR

Text to Speech

The voice your Arabic agent speaks with.

Arabic text to speech API. First audio in under 50 ms, 16 voices, Gulf dialects and MSA, built for calls that get interrupted.

Get your API keySee the agent plugins
Docs 
curl -N -X POST "https://api.munsit.com/api/v1/text-to-speech/faseeh-v1-preview" \
  -H "x-api-key: $MUNSIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "أهلاً بك، معك المساعد الصوتي. كيف أقدر أساعدك؟",
    "voice_id": "ar-najdi-male-2",
    "streaming": true
  }' --output speech.pcm

16 voices. Gulf dialects and MSA.

Pick the voice, pick the dialect, hear it before you ship it. No pause in the wrong place.

MajedClear, calm male voice with a steady, measured pace and a medium pitchEmiratiMale

Voice

Speed:1.00x
Speed1.00x
0.7x1.2x

Controls the speed of the generated speech. Higher values make the voice speak faster.

Need a brand voice? Cloning is consent-based and hosted in-region. Read the cloning guide

Works with your agent stack.

Drop-in plugins for the frameworks Arabic voice agents are actually built on. Nothing to wrap, nothing to proxy.

LiveKit

Munsit TTS as the TTS node in a LiveKit Agents pipeline.

livekit-plugins-munsit

Pipecat

Drop-in service class for Pipecat’s real-time pipeline.

pipecat-plugins-munsit

VAPI

Select Munsit TTS as the voice on any VAPI assistant.

voice: faseeh

Ultravox

Wire Munsit TTS in as the speech output for Ultravox calls.

tts.provider = faseeh

Talk over it. It stops, like a person would.

The caller interrupts, the voice stops in under 50 ms, the agent listens.

Time to first audio byte

Munsit TTS<50 ms
Typical global TTS

Stops instantly

The audio stops right away. No leftover words after the caller moves on.

Turn detection is built in

Munsit TTS knows when someone’s done talking in Arabic. No guessing from silence.

What Arabic agents actually do.

Someone calls in expecting an answer, not a script. That’s exactly what the ten biggest teams on Munsit TTS are building for.

Customer support

Answers questions, checks orders, and escalates when needed.

Sales & outbound

Qualifies, follows up, and collects, without reading a script.

Information delivery

Runs public info lines, in formal Arabic when needed.

Virtual assistants

Books, reminds, and guides, right inside your product.

Healthcare agents, with the compliance attached.

Appointments and reminders are the most common use of Munsit TTS. Regulated teams get in-region deployment and reviewer-ready controls from day one.

Talk to us about healthcare
Deployment & controls
In-regionOn-premPDPLNCA
Features
  • Appointment scheduling
  • Booking confirmation
  • Reminder calls
  • Pre-visit intake

Everything else you’ll need.

Munsit TTS and Munsit STT share one key, so giving an existing transcription integration a voice does not mean a second account or a second bill.

Streaming Synthesis
أهلاً بك، معك المساعد الصوتي. كيف أقدر أساعدك؟
audio out
chunk 01 · 12 KB186ms
chunk 02 · 14 KB271ms
chunk 03 · 11 KB348ms
WSS /websocket/text-to-speech
Mid-stream Cancellation
aborted at 340ms
AbortController / socket close
Arabic-aware Endpointing
silence
thresholdكيف أقدرأساعدك اليوم؟cuts early
Munsitكيف أقدر أساعدك اليوم؟turn end
built in
Dialect-native Voices
Fahadsaudi
Lamasaudi
Mansouremirati
+ MSA and the rest of the roster
GET /voices
Voice Cloning
reference · 30s
Lamaenrolled
POST /voices/clone
Register Control
colloquialشلونك؟ كيف أقدر أساعدك؟
msaكيف حالك؟ كيف يمكنني مساعدتك؟
per-request parameter
Output Encoding
PCM24 kHz · raw, lowest latency
WAV24 kHz · headered
WAV · PCM
Batch Synthesis
01done
02done
0364%
142 lines · off the real-time path
POST /text-to-speech/{model_id}

Two foundation models for everything you build.

Arabic speech-to-text and text-to-speech on one key, one base URL. Adding a voice to an existing integration is a new endpoint, not a new account.

Explore Speech-to-Text

Munsit STT

  • Real-time streamingPartials in about half a second, over WebSocket.
  • Dialect and registerColloquial and broadcast MSA in one request, no flags.
  • Word-level timestampsEvery word carries its own timing, streaming and batch alike.
  • Speaker diarizationLabels each speaker, which calls and sessions depend on.
See documentation

Munsit TTS

  • Under 50 ms to first byteFast enough to interrupt, and to be interrupted.
  • Clean cancellationAbort mid-sentence and generation stops. No orphaned audio.
  • 16 voicesSaudi, Emirati and Fusha, not a regional tint.
  • Arabic-aware endpointingKnows when the caller has finished, not just when they paused.
See documentation

What matters when the voice is in a conversation.

A live call can’t wait. Most text-to-speech was built for narration, not conversation, where nobody’s on the other end expecting a reply.

  • Munsit TTS
    Time to first audio byte
    <50 ms
    Cancellation mid-sentence
    Clean abort, generation stops
    Arabic-aware turn detection
    Built in
    Arabic dialects
    Native across 25+ dialects & MSA
    Deployment
    In-region, on-prem or on-device
    Voice cloning
    Arabic-tuned, expressive
  • Typical global TTS
    Time to first audio byte
    ~300–600 ms
    Cancellation mid-sentence
    Often finishes the utterance
    Arabic-aware turn detection
    None or basic
    Arabic dialects
    Generic Arabic accent
    Deployment
    Cloud-only
    Voice cloning
    English-optimised
What an agent needsMunsit TTSTypical global TTS
Time to first audio byte<50 ms~300–600 ms
Cancellation mid-sentenceClean abort, generation stopsOften finishes the utterance
Arabic-aware turn detectionBuilt inNone or basic
Arabic dialectsNative across 25+ dialects & MSAGeneric Arabic accent
DeploymentIn-region, on-prem or on-deviceCloud-only
Voice cloningArabic-tuned, expressiveEnglish-optimised

Give your product an Arabic voice.

Start speaking Arabic with one API key. Pick a voice and hear it for yourself.

Get your API keyRead the docs