kenpath

svara-TTS v1 — Open Multilingual TTS for India’s Voices [](https://huggingface.co/kenpath/svara-tts-v1) [](https://huggingface.co/spaces/kenpath/svara-tts) [](https://colab.research.google.com/drive/15YxFo1DzdQNbFUIZ1HJA4AN4oHqKxGtg) [](https://github.com/Kenpath/svara-tts-inference) svara-TTS is a developer-first multilingual TTS model for 19 languages (18 Indic + Indian English). Built on an Orpheus-style discrete audio token approach, it targets clarity, expressiveness, and low-latency on commodity GPUs/CPUs. It supports light-weight emotion/style control (e.g., ` `, ` `, ` `, ` `) and simple speaker identities (`Language (Gender)`), with zero-shot adaptation paths. - Languages (19): Hindi, Bengali, Marathi, Telugu, Kannada, Bhojpuri, Magahi, Chhattisgarhi, Maithili, Assamese, Bodo, Dogri, Gujarati, Malayalam, Punjabi, Tamil, Nepali, Sanskrit, Indian English. - Expressivity: End-of-utterance style tags; natural prosody; code-switch aware. - Latency & Deployment: Works well with GGUF exports; suitable for edge/CPU scenarios. - Adaptability: LoRA-friendly for quick speaker/domain specialization. Try it live on the Demo Space, or on Colab Deployment scripts and inference repo will be available soon. Watch our Github for updates - Place style/emotion tags at the end of the sentence: `आज... सच में अच्छी खबर है — शाम को मिलते हैं! ` - Use punctuation to hint prosody (ellipses, commas, exclamation). - For technical or dense text, end with ` ` to prioritize intelligibility. > Speaker IDs follow a simple convention: `Language (Gender)` (e.g., `Marathi (Male)`). Trained on 2000+ hours of open, high-quality speech from SYSPIN, RASA, IndicTTS, and SPICOR, covering ~50 speakers (balanced male/female) across 19 languages. Data was curated to encourage natural prosody, broad coverage, and stable multilingual transfer. See Acknowledgments for provenance. - Multilingual assistants, IVR, learning apps, reading aids, accessibility tools - Content localization (education, public-information, civic services) - Research on Indic prosody, emotion control, cross-lingual transfer - Impersonation of private individuals or public figures without consent - Deceptive content (fraud, harassment, misinformation) - Safety-critical deployments without human oversight - Proper nouns & rare entities: may require spelling hints or ` `. - Very long sentences: chunk or add punctuation for natural prosody. - Emotion strength: varies by language due to data density. - Code-mixing: common patterns work; it’s not a deterministic rules engine. Many of these improve with targeted LoRA finetuning and better preprocessing. By using this model, you agree to follow applicable laws and ethical guidelines. Avoid impersonation, harassment, targeted deception, or other harmful uses. Where appropriate, disclose synthetic speech to end users. - Model: https://huggingface.co/kenpath/svara-tts-v1 - Demo Space: https://huggingface.co/spaces/kenpath/svara-tts - Inference repo: https://github.com/Kenpath/svara-tts-inference - Colab: https://colab.research.google.com/drive/15YxFo1DzdQNbFUIZ1HJA4AN4oHqKxGtg This work was developed by Kenpath Technologies for the open-source community. We also thank RunPod for the startup credits that supported our GPU compute. - Canopy Labs — Orpheus: foundational ideas & open release Release: https://canopylabs.ai/releases/orpheuscanspeakanylanguage - SPIRE Lab, IISc Bangalore — SYSPIN (multilingual studio) and SPICOR (Indian English) - AI4Bharat — RASA expressive speech - IIT Madras — IndicTTS - Unsloth — helpful notes & tooling - RunPod — startup GPU credits that accelerated experiments

Files - bharat-pii-gemma-3-1b-it-v0.3-f16.gguf - bharat-pii-gemma-3-1b-it-v0.3q80.gguf Notes - GGUF is self-contained (tokenizer/config embedded). - Converted with llama.cpp. Suggested: q80 for general use, f16 for reference.

Files - bharat-pii-gemma-3-270m-it-v0.7-f16.gguf - bharat-pii-gemma-3-270m-it-v0.7q80.gguf Notes - GGUF is self-contained (includes tokenizer/config metadata). - Converted with llama.cpp. - Suggested: q80 for general use, f16 for reference/evaluation. Example (llama.cpp) - ./main -m ./bharat-pii-gemma-3-270m-it-v0.7q80.gguf -p "My name is ..."

- Developed by: kenpath - License: apache-2.0 - Finetuned from model : unsloth/Qwen3-4B-Instruct-2507 This qwen3 model was trained 2x faster with Unsloth and Huggingface's TRL library.