Rumik AI

Rumik AI is an Indian research lab releasing Rumik OSS 1, an open source pre-trained voice model that speaks more than 20 languages and ships with fully open weights. The lab is pitching it as India's first state-of-the-art open source voice foundation model, trained on roughly 66,000 hours of audio and evaluated against closed frontier models on benchmarks that include emotion realism. This launch matters because most competitive expressive TTS systems remain proprietary and API-gated, so a downloadable checkpoint that covers Indian and global languages gives developers a base to fine-tune, self-host, and build voice products on without per-minute vendor pricing. The release fits into Rumik's broader work on speech generation, which so far has centered on the Silk family of models. Its earlier research post describes a transformer backbone TTS model that treats speech generation like next-token prediction, using a language model to output discrete audio codes that are then decoded back to waveforms through a latent encoder-decoder , and the team has continued to publish work on voice parameterization by pace, accent and tone with support across 22 languages . OSS 1 extends that lineage into a weights-available artifact aimed at the open source community rather than a hosted API. The company is led by founder and CEO Rohan Chaudhary, who posts as @lets_dig_deeper, with Yogesh Maheshwari as co-founder and CTO . Rumik describes itself on its site as a research lab building foundation models across text, voice and video, and this launch is a signal of intent to compete at the model layer from India rather than only wrap existing providers. Founders and operators evaluating voice stacks, multilingual agents, or India-language coverage should treat OSS 1 as a base worth benchmarking against Eleven Labs, Sesame, and other closed incumbents before committing to a vendor.