Our models

One family. One mission. Arabic voice, finally native.

Three TTS tiers · Saree' for streaming speed, Faseeh for premium narration with non-verbal tags, Mulhim for voice design from text · plus Hakim Arab v2 for dialect-aware transcription. Built from the ground up for Arabic, fluent in 30+ other languages.

TTS · Speed tier

Hakim Saree' v1.3

Streaming TTS

The speed tier of the Hakim TTS family. A streaming decoder tuned for low first-byte latency across 30+ languages. The default for voicebots, IVR, and any path where latency is the binding constraint.

Languages
30+ supported
Arabic dialects
15 covered
First chunk
≤150 ms
Sample rate
24 kHz native
  • Usage-based pricing. No per-seat fees.
  • Commercial use on every paid tier.

TTS · Premium quality

Hakim Faseeh v2

Expressive narration + tags

The premium-quality tier · built for narration, podcasts, and branded ads where naturalness matters more than the lowest latency. Honours the full non-verbal tag set ([laugh], [breath], [whisper], …) across 41 languages.

Languages
41 supported
Arabic dialects
15 covered
Non-verbal tags
Full token set
Sample rate
24 kHz native
  • Premium tier pricing.
  • Commercial use on every paid tier.

TTS · Generative

Hakim Mulhim v3

Most capable
Voice design from text

The full capability set · premium quality plus voice generation from a natural-language description. Pair voice_prompt with any voice for a tonal overlay, or design a fresh timbre from text and save it to your library. 41 languages.

Languages
41 supported
Voice design
From prompt
Non-verbal tags
Full token set
Sample rate
24 kHz native
  • Premium tier pricing.
  • Commercial use on every paid tier.

A multi-dialect transcription model that handles MSA and the major Khaleeji / Levantine / Maghrebi variants without manual switching. Returns punctuated output with optional diarisation, batch + streaming modes, and long-form audio up to 60 minutes per request.

Languages
30+ supported
Arabic dialects
15 covered
Diarisation
Built-in
Long-form audio
Up to 60 min / request
  • Per-minute pricing. No per-seat fees.
  • Commercial use on every paid tier.

Transcription

Hakim Arab v2

Dialect-aware Arabic speech recognition.

Dialect coverage

Every major Arabic dialect, end to end.

Fifteen Arabic dialect varieties grouped into eight families, the same registry the SDK reads from. Every dialect is GA on every capability · TTS, STT, voice cloning, streaming.

Arabic dialect coverage across the Hakim TTS family (Saree' / Faseeh / Mulhim), Hakim Arab v2 STT, voice cloning, and streaming delivery.
DialectTTSSTTCloningStreaming
Modern StandardMSAtts: availablestt: availablecloning: availablestreaming: available
KhaleejiGulftts: availablestt: availablecloning: availablestreaming: available
EgyptianMasritts: availablestt: availablecloning: availablestreaming: available
LevantineShaamitts: availablestt: availablecloning: availablestreaming: available
MaghrebiNorth Africantts: availablestt: availablecloning: availablestreaming: available
IraqiMesopotamiantts: availablestt: availablecloning: availablestreaming: available
SudaneseEast-African Arabictts: availablestt: availablecloning: availablestreaming: available
YemeniPeninsular Arabictts: availablestt: availablecloning: availablestreaming: available
  • Available

Hear every dialect

Same TTS family. Every dialect.

Press play on any card below. Same Hakim TTS family, same prompt, different dialect, so your ears, not a benchmark, decide. All three tiers (Saree' / Faseeh / Mulhim) inherit the full 15-dialect catalogue.

Qualitative samples only, no numeric accuracy claims until we publish a peer-reviewed methodology.

Modern Standard

narrative
Modern Standard· layla-msa
0:000:00

Khaleeji

conversational
Khaleeji· reem-khaleeji
0:000:00

Egyptian

conversational
Egyptian· yusuf-egyptian
0:000:00

Levantine

social-media
Levantine· nour-levantine
0:000:00

Maghrebi

conversational
Maghrebi· amir-maghrebi
0:000:00

Iraqi

narrative
Iraqi· omar-iraqi
0:000:00

Sudanese

conversational
Sudanese· salma-sudanese
0:000:00

Yemeni

On the roadmap

We ship samples once a dialect clears our native-reviewer bar. This one is still in training.

Beyond Arabic

30+ languages on the baseline tier, 45 on Mulhim v3.

Saree' v1.3 ships 32 supported languages, Faseeh v2 widens to 41, Mulhim v3 to 45 (see the per-tier counts on the cards above). Every code below ships on the Saree' GA tier · Faseeh and Mulhim inherit each language Saree' already supports.

General availability

Available on every paid tier, with SLA coverage on the p95 latency published above.

  • English
  • French
  • Spanish
  • German
  • Italian
  • Portuguese
  • Turkish
  • Urdu
  • Hindi
  • Persian
  • Hebrew
  • Dutch
  • Russian
  • Polish
  • Japanese
  • Korean
  • Chinese
  • Thai
  • Vietnamese
  • Indonesian
  • Malay
  • Swahili
  • Greek
  • Finnish
  • Swedish
  • Norwegian
  • Danish
  • Burmese
  • Khmer
  • Lao
  • Filipino

Ready to ship?

Start free. See it for yourself in 30 seconds.

5,000 characters and 10 minutes of transcription on the house. No card. Production APIs from the first call.

Model FAQ

Picking the right tier and language.

Common questions about coverage, latency, and how the tiers differ.