Promote Sign in Sign up

Audio model

Rumik OSS 1

Rumik OSS 1 is an open source text to speech model built for Indic languages, generating expressive, natural sounding speech across 22 languages with native transcriptions. It supports code switching, inline emotion tags, and vocalization effects like laughter and sighs. Built on a Tiny Aya Fire text foundation with a MIMI audio codec, it uses a decoder only transformer to predict 24kHz audio tokens. Voice output can be controlled by pace, accent, and tone.

Rumik Audio Open weights cc-by-nc-4.0
Visit model website ↗ Hugging Face
Type
Audio
Weights
Open
Downloads
2k
Likes
60
Pipeline
text to speech
License
cc-by-nc-4.0

Model hub

rumik-ai/rumik-oss-1

Repo
rumik-ai/rumik-oss-1
Library
transformers
Access
Public
Technical tags and languages
hi bn ta te mr gu kn ml pa or as ur ne transformers safetensors rumik_oss text-generation tts text-to-speech indic expressive mimi speech custom_code
Open on Hugging Face

More from Rumik

4

About Rumik

1

Rumik is an Indian AI research lab building emotionally intelligent AI companions. Flagship product Ira is a multilingual AI being that understands intent, infers emotions, and converses in Hinglish, Bangla, and Marathi — available on WhatsApp, Telegram, and iOS/Android. Three proprietary models: Silk-1 (expressive voice), Mesh (persistent memory), and Peek (contextual understanding). Founded April 2024. Raised $6.5M from Info Edge, Huddle Ventures, and Elevation Capital. CEO: Rohan Chaudhary.