Every synthesis request names a voice through the voice field. Voices differ in dialect, timbre, apparent age, and delivery style, the choice sets the character of the output more than any other parameter.
Fetch the catalog programmatically with GET /v1/voices so your application stays current as voices are added.

Arabic voices

The core of the catalog. Najdi leads because it is the accent this product is built around.

Najdi, Saudi Arabia

Hijazi, Saudi Arabia

The west coast register: Jeddah, Makkah and Madinah. A Jeddah caller hears a Najdi agent as out of town, so a west coast line should answer in Hijazi.

Omani, Oman

Muscat and the interior. Omani is neither Saudi nor Emirati, and the words give it away.
The voice carries the sound. The words carry the dialect. A Najdi agent says الحين, وش and أبشر; a Hijazi one says دحين, إيش and إيوه; an Omani one says شو, زين and ما عليه. Put the dialect in the agent’s instructions as well as choosing the voice.

Gulf, MSA, and Egyptian

When you do not know which dialect your callers speak, yousef or khalid in Modern Standard Arabic is the safe default. MSA reads as formal but is understood everywhere; a mismatched regional dialect sounds foreign.

English voices

Dialects

Filter the catalog by any of these:

Styles

Each voice carries one or more delivery styles. Match the style to the line rather than to preference.

Professional

Service desks, support lines, anything where the caller has a problem.

Formal

Government-facing, compliance, disclosures, announcements.

Casual

Everyday conversation, consumer-facing lines.

Energetic

Sales, outbound campaigns, promotions.

Choosing a voice

Service and support lines read better with an unhurried delivery: layla or nouf in Arabic, clementine or vespera in English. Formal and official lines want faisal or khalid. Sales and outbound can carry the pace of omar, astra, or cupola.
Test with callers who speak the way your customers actually speak, not colleagues reading a script. Accent coverage is the most common source of early failures, and it shows up within a handful of real calls.

Auditioning quickly

GET /v1/voices/{id} returns a ready-to-send sample_request in the right language for that voice, post it straight to /v1/speech to hear the voice without writing anything.