OpenAI TTS voices 2026 — 13 options vs EasyVoice's 66-voice catalog

How many voices does EasyVoice have in 2026?

EasyVoice has 66 voices: 56 multilingual voices powered by the Kokoro-82M model across 8 languages, and 10 Arabic-language voices powered by the Supertonic model — 66 voices across 9 languages total. The 56 Kokoro voices cover American English (28 voices: 15 female af_ + 13 male am_), British English (10 voices: 5 female bf_ + 5 male bm_), Spanish (3), French (1), Hindi (4), Italian (2), Japanese (5), and Portuguese (3). The 10 Supertonic voices are dedicated Arabic MSA (Modern Standard Arabic) voices — male and female — from a separate purpose-built Arabic TTS engine rather than a multilingual model with Arabic capability bolted on.

OpenAI's 13 voices are one shared set used for every language it supports, and OpenAI describes them as optimized for English; there is no voice built specifically for Arabic, Japanese, or Hindi. EasyVoice ships voices built for each language instead. Two caveats keep this honest: Italian and Arabic are in beta on EasyVoice (Arabic: "A newer engine that has not yet passed our listening panel."), and Spanish, French, Hindi, Italian, Japanese and Portuguese (Brazil) voices are Pro-only — the free plan covers American English, British English and 2 Arabic voices.

The 2026 EasyVoice voice catalog — language and count matrix

By language and voice count — generated from the live voice list, so the numbers always match production. The 56 Kokoro-82M multilingual voices span 8 languages (American English carries the deepest bench at 28 voices across neutral, warm, energetic, and authoritative registers), and the 10 Arabic MSA voices run on a dedicated Supertonic engine. OpenAI lists 13 voices (9 on tts-1 and tts-1-hd), none of them language-specific:

OpenAI voice to EasyVoice mapping — 2026 catalog

If you're migrating from OpenAI TTS, the full voice-by-voice mapping table — tone notes, use cases and sample scripts — lives on the OpenAI TTS voices page (/openai-tts-alternative/voices). In short, 6 of OpenAI's voices have a direct EasyVoice match: alloy → af_alloy, echo → am_echo, fable → bm_fable, onyx → am_onyx, nova → af_nova, shimmer → bf_lily. OpenAI's other voices (ash, ballad, coral, sage, verse, marin and cedar) are not mapped yet — pick by tone from the full catalog.

Beyond the 6 direct equivalents, EasyVoice's catalog includes voices without an OpenAI analog: af_heart (warm American female, optimized for long-form narration), am_adam (authoritative American male), bm_lewis (contemporary British male), and more across all supported languages. A developer migrating from OpenAI starts with the direct matches and then has access to 60 additional options to optimize for tone, register, and language.

Arabic MSA — 10 dedicated voices vs OpenAI's generalized approach

The 10 Arabic voices in EasyVoice's catalog are powered by a separate synthesis engine (Supertonic) built for Modern Standard Arabic. The voices are labeled ar_m1 through ar_m5 (male) and ar_f1 through ar_f5 (female); ar_m1 and ar_f1 are on the free plan, and ar_f2, ar_m3, ar_f3, ar_m4 and ar_f4 are available through the API but hidden from the studio's voice picker. These are not Kokoro-82M voices with an Arabic language code — they are a different model architecture. Arabic is in beta on EasyVoice — a newer engine that has not yet passed our listening panel.

OpenAI's TTS API has no Arabic-specific voice; the same shared voices read every language it supports, and OpenAI describes them as optimized for English. Whether that is good enough depends on your listeners, so test both with native speakers on your own scripts. For Arabic IVR, Arabic content narration, or Arabic customer service, a purpose-built Arabic engine is the more targeted choice — with the beta caveat above.

How often is the EasyVoice voice catalog updated?

The catalog on this page is generated from EasyVoice's live voice list, so its counts always match production. Voices are added when new Kokoro model weights or new language-specific engines are ready; languages whose engine has not yet passed our listening panel ship labelled beta (currently Italian and Arabic). The Arabic Supertonic voices were added as a dedicated engine (replacing the earlier best-effort Arabic output from Kokoro) after an internal Arabic bake-off measured Supertonic's word error rate on MSA.

Switch to EasyVoice if you need Arabic, Japanese, Hindi, Portuguese, French, Italian, or Spanish voices built for each language (Pro, apart from 2 free Arabic voices; Italian and Arabic are beta), or if voice breadth and tonal variety matter for your application — 66 voices across 9 languages give you per-use-case options that one shared set of 13 cannot. Stay on OpenAI if English is your only language and you need one of OpenAI's specific voices with no migration cost — the direct EasyVoice equivalents are close to, but not the same as, OpenAI's proprietary voice models.