SILMA AI

SILMA builds Arabic-first speech models. Its hosted text-to-speech service is described as dialect-aware and its app offers Modern Standard Arabic and a "Saudi dialect" option; it also publishes an open first-generation model on Hugging Face under the Apache 2.0 licence for Arabic and English, unusual among the vendors in our guide. Two documented details matter for Arabic users. First, the model accepts diacritics in the input and applies automatic vocalisation when they are missing, which is where most Arabic voices stumble. Second, SILMA is one of the fifteen or so systems in Navid's Arabic TTS Arena, the blind listening comparison launched in Riyadh in March 2026, although the arena has published no rankings yet. We found no documentation of where it is headquartered, where audio is processed, or of voice cloning consent rules. Pricing in November 2026: from $0.020 per minute of generated audio; we found no free tier documented.