SILMA TTS v1 is a 150 million parameter Arabic/English bilingual Text-to-Speech (TTS) model reportedly released by SILMA AI on March 15, 2026. It synthesizes input sentences into natural-sounding speech and supports instant voice cloning, reflecting the speaker's tone using only reference audio of less than 8 seconds. While large-scale general-purpose voice models pack multiple languages and features into a single, heavy system, SILMA TTS v1 is closer to a small engine focused on a clear language scope of Arabic and English. In particular, it supports Arabic diacritics and text normalization, allowing it to process input sentences into a form suitable for pronunciation in Arabic speech synthesis, where there is a large gap between the written form and the actual pronunciation. In existing Arabic TTS workflows, the reading style of numbers, abbreviations, and symbols, diacritics that supplement vowel information, and the differences between dialects and standard language directly affect the quality of the synthesis. Simply applying an English-centric model can lead to instability in Arabic-specific phonemes or prosody, and models that require long recordings or separate fine-tuning for high-quality speaker adaptation can be burdensome for rapid experimentation and deployment. The key differentiator of SILMA TTS v1 is that it provides both Arabic/English synthesis and short reference audio-based cloning within a relatively lightweight 150M parameter model. Similar to how GPT generates new text based on short prompts, this model is designed to use audio examples of less than 8 seconds as clues to the speaker's characteristics, allowing it to read new sentences in a voice of the same style. If the Apache-2.0 license is confirmed in the official repository, it becomes an open-source option that can be used not only for research but also for prototyping commercial services. From the perspective of life science researchers, it can be used for multilingual research guides, patient education materials, and accessible content creation. For example, procedures for participating in clinical trials or precautions before sample collection can be prepared in Arabic and English, and consistent voice tones for both languages can be created using less than 8 seconds of reference audio from an approved speaker. By synthesizing text that has applied text normalization and Arabic diacritics, and then performing native speaker review and back-transcription, an evaluation pipeline can be established to check expressions with high error costs, such as numbers, dosage units, and test schedules. However, this model is not a medical device, and expert review and the institution's consent and personal information protection procedures are required for clinical instructions or content intended for patients. In educational and research communication, it can be used to convert abstracts, laboratory safety training, and bioinformatics tutorials into Arabic and English audio materials. The characteristic of being a 150M parameter model suggests the possibility of experimentation in limited computational environments compared to large voice models, but the actual latency and memory usage should be measured separately in the official documentation and on the target hardware. When using voice cloning, explicit consent from the reference speaker must be obtained, and the synthesized voice must be indicated. It is also safe to use a pronunciation dictionary or human-verified input notation in conjunction with technical terms such as disease names, gene names, and drug names.

SILMA TTS v1 is a 150 million parameter Arabic-English bilingual Text-to-Speech (TTS) model reportedly released by SILMA AI on March 15, 2026. It synthesizes input sentences into natural-sounding speech and supports instant voice cloning, reflecting the speaker's tone using only reference audio of less than 8 seconds. While large-scale general-purpose voice models pack multiple languages and functions into a single, heavy system, SILMA TTS v1 focuses on a smaller engine with a clear language scope of Arabic and English.