Speech to Text
Word-level transcripts in 25+ dialects, with the English words left as English.
Build with Arabic voice AI
that understands
Speech to text ranked #1 on six Arabic benchmarks.
Text to speech, first audio in under 50 ms. One key, one API call.
MajedClear, calm male voice with a steady, measured pace and a medium pitchEmiratiMale
Speed:1.00x
Speed1.00x
0.7x1.2x
Controls the speed of the generated speech. Higher values make the voice speak faster.
Trusted by 250+ companies and government agencies
كيف حالك؟ كل شيء على ما يُرام
0:00 / 0:04
Text to Speech
Arabic in the dialect you pick, first audio in under 50 ms.
Real-time streaming
Partial words as people speak, final text on the pause. One WebSocket.
Every dialect. One call.
No dialect flag to set. Send the audio and it detects dialect automatically.
Made for developers
REST and streaming APIs, an on-device SDK, and native LiveKit plugins. You can copy over the script below to try it.
curl -X POST "https://api.munsit.com/api/v1/audio/transcribe" \ -H "x-api-key: $MUNSIT_API_KEY" \ -F "file=@meeting.mp3" \ -F "model=munsit"
Response
"data": { "transcription": "اجتماع الفريق يبدأ الساعة العاشرة", "duration": 14.2}
STT & TTS endpoints
REST and streaming APIs for Arabic speech-to-text and text-to-speech.
Runs fully on-device
No audio ever leaves the phone or laptop. Native SDKs for iOS, Android, macOS, Windows, and Linux.
LiveKit integration
Go live fast with native LiveKit and plugin integrations for real-time Arabic voice applications.
Scaling real
Arabic voice use cases
Explore where Munsit powers better Arabic voice experiences.
Automate Arabic customer calls
Voicebots that understand real speech and reply instantly.
Turn calls & meetings into records
Send the audio, get a diarized transcript with speakers and action items.
Caption and index media at scale
Live captions for broadcast and events, searchable archives for everything recorded.
Add voice to your app
Dictation, voice commands, and voice search in the dialect your users actually speak.
Subtitles, dubbing and voice-over
A script to Arabic audio in your dialect, a video dubbed, subtitles exported.
Index Arabic media archives
Turn news, broadcasts, & calls into searchable data: sentiment, keywords, and more.
Need the full agent, not just the voice? Munsit is the Arabic voice layer inside AGNTIX, CNTXT AI’s agent platform.Explore AGNTIX
Number one on
six Arabic benchmarks.
18% average word error rate across SADA, Common Voice 18.0, MASC clean and noisy, MGB-2 and Casablanca. Values as the leaderboard shows them.
Model Accuracy %Higher is better
- Munsit82.00
- Transcribe74.14
- NVIDIA65.26
- Deepgram Nova-364.14
- OpenAI Whisper63.15
- Meta Seamless M4T61.84
- ElevenLabs Scribe59.95
- OpenAI ChatGPT55.03
- Microsoft Azure54.27
What matters in a live call
| Munsit TTS | Typical global TTS | |
|---|---|---|
| Time to first audio byte | <50 ms | ~300–600 ms |
| Cancellation mid-sentence | Clean abort, generation stops | Often finishes the utterance |
| Arabic-aware turn detection | Built in | None or basic |
| Arabic dialects | Native across 25+ dialects & MSA | Generic Arabic accent |
| Deployment | In-region, on-prem or on-device | Cloud-only |
| Voice cloning | Arabic-tuned, expressive | English-optimised |
- 18%
- average WER, #1 on six Arabic benchmarks
- 25+
- Arabic dialects, plus Arabic–English code-switching
- 250+
- companies and government agencies
- 300,000
- people use Munsit
Every dialect is the same call.
Arabic isn’t one language. A Moroccan and a Kuwaiti sound different, but most AI hears the same generic Arabic. Munsit hears the dialect on its own. Pick one, play it, see the reply.
شحالك؟ كل شي زين؟
sh-ḥālak? kill shay zēn?
POST /audio/transcribe -F "file=@clip.mp3" -F "model=munsit"
Subscription.
Credits that never expire.
Start free with 10,000 credits, no card required.
10K credits2× concurrency
200k credits / mo2× concurrency
1M credits / mo3× concurrency
3M credits / mo5× concurrency
1,000 credits ≈ 1 minute of transcription or 500 characters of speech.See all plans
Build with Arabic from day one.
One API key. Speech-to-text and text-to-speech built for Arabic. Try it for free.