Colloquial (Saudi)
شلونك؟ كل شي تمام؟
shlōnak? kill shay tamām?
Transcribed as spoken, not textbook MSA.
Speech to Text
The most accurate Arabic speech to text API. Colloquial dialect, broadcast MSA and Arabic-English code-switching. One request, no dialect flag.
import websockets, json
ws = await websockets.connect(
"wss://api.munsit.com/api/v1/listen"
"?model=munsit-en-ar&encoding=linear16&sample_rate=16000",
additional_headers={"x-api-key": MUNSIT_API_KEY},
)
async for chunk in microphone(): # 16 kHz PCM
await ws.send(chunk)
await ws.send(json.dumps({"type": "CloseStream"}))583k real transcriptions, 99.8% of them completed successfully, as of July 2026.
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.
Munsit handles colloquial speech, broadcast MSA and Arabic-English code-switching in the same request. No dialect parameter, no per-region endpoint.
# the same call produces all three results below curl -X POST "https://api.munsit.com/api/v1/audio/transcribe" -H "x-api-key: $MUNSIT_API_KEY" -F "file=@clip.wav"
شلونك؟ كل شي تمام؟
shlōnak? kill shay tamām?
Transcribed as spoken, not textbook MSA.
وفي ختام النشرة، نستعرض أبرز العناوين
wa-fī khitām an-nashra, nastaʿriḍ abraz al-ʿanāwīn
Formal register, ready to index.
حابب أعمل deploy على الـ staging الأول
ḥābib aʿmil deploy ʿala l-staging al-awwal
One clean transcript. Names, numbers, dates and the English words your customers use come back the way they said them.
News archives, media libraries and back-catalogues turned into searchable text, with word-level timestamps so every hit points to a moment.
Post files or point at storage, up to 500MB or 60 minutes each.
POST /audio/transcribeConcurrency is an account limit, not a code change. Throughput scales with the key, not the integration.
documented per-plan limitsWord-level timings on every result, so search returns the second a phrase was said, not just the file.
JSON · SRT · TXT · DOCXArabic speech to text with diarization, timestamps, translation and export formats, each one endpoint on the key you already have.
Most of the audio Munsit sees is under five seconds. Short commands get a fast path, so you get text back almost as soon as you speak.
Dictate into any field, search box, or form, as fast as you can speak.
Short commands turned into text fast enough to act on, in any dialect or MSA.
Streaming results let your agent start responding before the caller stops talking, not after they pause.
Drop Munsit into the framework you’re already using, then give the same agent a voice with Munsit TTS. Speech in, speech out, one API key.
Use Munsit as the STT node in your LiveKit Agents pipeline.
livekit-plugins-munsitDrop Munsit in as a service class for Pipecat’s real-time pipeline.
pipecat-plugins-munsitSelect Munsit as the transcriber for any VAPI assistant.
transcriber: munsitWire Munsit in as the speech input for Ultravox calls.
stt.provider = munsitArabic 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.
Start transcribing Arabic with one API key. Live or uploaded, it’s ready to test.