Meeting notes
Multi-speaker diarisation, built in.
Get labelled speakers and punctuated transcripts for Arabic + English meetings without stitching tools together.
Speech-to-text
Arabic transcription that actually understands dialect.
Hakim Arab v2 transcribes MSA + 14 regional Arabic dialects with dialect-aware decoding, speaker diarisation, and code-switching for bilingual audio.
How it works
From audio to text in three calls.
The same pipeline that powers our enterprise customers, surfaced through a single endpoint.
Upload the audio
POST a file up to 100 MB (
wav,mp3,flac,m4a,webm) with an optional language hint.We transcribe
Hakim Arab v2 detects dialect, applies punctuation, and diarises speakers in a single pass.
Get structured text
Return plain text, verbose JSON with word timings, or SRT/VTT, one response format flag.
Try it
Drop a file in the playground.
Anonymous uploads aren't supported yet, create a free account to transcribe audio in a few clicks.
Signup required
The STT anonymous endpoint is deferred, sign up to drop a file, pick a language, and get a transcript in the playground.
Where teams ship this
Transcription that survives real Arabic audio.
Built for dialect-heavy, code-switched, and noise-laden recordings, not just studio reads.
Media & captions
SRT and VTT in one request.
Caption dialect-specific interviews, news segments, and social clips with word-level timings.
Contact centre
Call transcripts for QA and analytics.
Feed call recordings straight into your analytics stack, dialect-aware Arabic keeps the signal intact.
Technical specs
Shape and limits at a glance.
- Model
- Hakim Arab v2Arabic-first streaming STT.
- Latency p95
- 90 msFirst-token, real-time mode; rolling 24-hour p95 across our four live regions.
- Languages
- 30+ supported · 15 Arabic dialects.MSA, Khaleeji, Egyptian, Levantine, Maghrebi, Iraqi, Sudanese, Yemeni.
- Modes
- Batch · real-time streaming.One model, two transport surfaces.
- Input formats
- wav · mp3 · flac · m4a · webm.Up to 100 MB per file.
- Diarisation
- Speaker labels returned per segment.Supports 1–8 speakers per track.
API
One upload, one call, one transcript.
Audio goes up, JSON comes back. Python-first because that's where most ingestion pipelines live.
See the full reference in the docs.
python
from hakim import Hakim
client = Hakim() # reads HAKIM_API_KEY from the environment
with open("call.wav", "rb") as audio_file:
transcription = client.audio.transcriptions.create(
model="hakim-arab-v2",
file=audio_file,
language="ar",
response_format="verbose_json",
)
print(transcription.text)
Ready to ship?
Start transcribing today.
Free tier includes 5 minutes of STT per month. Move to Starter when the load outgrows it.