README.md

---
license: cc-by-nc-4.0
language:
- hi
- bn
- te
- ta
- mr
- gu
- kn
- ml
- pa
- or
- as
- ur
- sa
- ne
- kok
- mai
- brx
- doi
- ks
- mni
- sat
- sd
- en
tags:
- audio
- text-to-speech
- tts
- piper
- vits
- mms
- onnx
- indic
pipeline_tag: text-to-speech
---

# BharatBhasha AI - Text-to-Speech (TTS) Models Catalog

## 📌 Repository Purpose & Motivation
This repository provides dedicated neural voice synthesis models, VITS checkpoints, and phonetic assets supporting natural offline voice synthesis across **all 22 Official Scheduled Indian Languages + English**.

### Why this repository exists:
1. **Un-gated Direct Distribution**: Eliminates gating screens, Hugging Face login prompts, or token requirements so mobile (Android) and desktop apps can fetch voices seamlessly.
2. **Hybrid Multi-Engine Architecture**: Provides both dedicated lightweight ONNX models (Piper Neural Voices, ~60 MB) for major languages, and high-coverage VITS models (Meta MMS, ~140 MB) for regional languages.
3. **Phonetic Independence**: Bundles the complete `espeak-ng-data` database and phonetic dictionaries for regional languages (e.g. Sanskrit, Bodo, Santali) so apps operate 100% offline without external system dependencies.

---

## 🏗️ Repository Folder Structure

```text
remiai3/TTS_MODELS_FOR_APP/
├── .gitattributes
├── README.md                                          # Documentation & Catalog
│
├── espeak-ng-data/                                    # Universal Phonetic Rules & Tables (17 MB)
│   ├── phontab, phonindex, phondata
│   └── voices/ (all phonetic tables)
│
├── [Piper Neural Voice ONNX Models & Configs]         # Ultra-Fast ONNX Voices (~60 MB each)
│   ├── hi_IN-pratham-medium.onnx (.json)              # Hindi
│   ├── te_IN-maya-medium.onnx (.json)                 # Telugu
│   ├── bn_BD-google-medium.onnx (.json)               # Bengali
│   ├── en_US-lessac-medium.onnx (.json)               # English
│   ├── mr_IN-google-medium.onnx (.json)               # Marathi
│   ├── ml_IN-arjun-medium.onnx (.json)                # Malayalam
│   ├── ne_NP-google-medium.onnx (.json)               # Nepali
│   └── ur_PK-fasih-medium.onnx (.json)                # Urdu
│
├── [Meta MMS VITS Models]                             # VITS Neural Voice Checkpoints (~140 MB each)
│   ├── mms_tts_tam/ (Tamil)                           # model.safetensors, config.json, vocab.json
│   ├── mms_tts_asm/ (Assamese)                        # model.safetensors, config.json, vocab.json
│   ├── mms_tts_guj/ (Gujarati)                        # model.safetensors, config.json, vocab.json
│   ├── mms_tts_kn/ (Kannada)                          # model.safetensors, config.json, vocab.json
│   ├── mms_tts_or/ (Odia)                             # model.safetensors, config.json, vocab.json
│   ├── mms_tts_doi/ (Dogri)                           # model.safetensors, config.json, vocab.json
│   └── mms_tts_mai/ (Maithili)                        # model.safetensors, config.json, vocab.json
│
└── [Indic Voice Phonetic Dictionaries]                # Lexical and phonetic assets
    ├── voice_tts_brx/ (Bodo)                          # model_info.json, brx_dict
    ├── voice_tts_kok/ (Konkani)                       # model_info.json, kok_dict
    ├── voice_tts_mni/ (Manipuri)                      # model_info.json, mni_dict
    ├── voice_tts_sa/ (Sanskrit)                       # model_info.json, sa_dict
    ├── voice_tts_sat/ (Santali)                       # model_info.json, sat_dict
    └── voice_tts_sd/ (Sindhi)                         # model_info.json, sd_dict
```

---

## 🌐 Direct Download Endpoints for Android & Mobile Apps
Clients can download individual voices on demand:

### 1. Piper Neural ONNX Voices & JSONs
```text
https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/{VOICE_FILENAME}.onnx
https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/{VOICE_FILENAME}.onnx.json
```
*Examples:*
- `https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/hi_IN-pratham-medium.onnx`
- `https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/te_IN-maya-medium.onnx`

### 2. Meta MMS VITS Models
```text
https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/{MODEL_DIR}/model.safetensors
https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/{MODEL_DIR}/config.json
```
*Example for Tamil:*
- `https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/mms_tts_tam/model.safetensors`
- `https://huggingface.co/remiai3/TTS_MODELS_FOR_APP/resolve/main/mms_tts_tam/config.json`

---

## 📊 Complete TTS Voice Inventory
| Language Code | Language | Voice Engine | Model Filename / Directory | Size | RAM Overhead |
| :---: | :--- | :--- | :--- | :---: | :---: |
| `hi` | Hindi | Piper ONNX | `hi_IN-pratham-medium.onnx` | 60 MB | < 100 MB |
| `te` | Telugu | Piper ONNX | `te_IN-maya-medium.onnx` | 60 MB | < 100 MB |
| `bn` | Bengali | Piper ONNX | `bn_BD-google-medium.onnx` | 73 MB | < 100 MB |
| `en` | English | Piper ONNX | `en_US-lessac-medium.onnx` | 60 MB | < 100 MB |
| `mr` | Marathi | Piper ONNX | `mr_IN-google-medium.onnx` | 73 MB | < 100 MB |
| `ml` | Malayalam | Piper ONNX | `ml_IN-arjun-medium.onnx` | 60 MB | < 100 MB |
| `ne` | Nepali | Piper ONNX | `ne_NP-google-medium.onnx` | 73 MB | < 100 MB |
| `ur` | Urdu | Piper ONNX | `ur_PK-fasih-medium.onnx` | 60 MB | < 100 MB |
| `ta` | Tamil | Meta MMS VITS | `mms_tts_tam` | 138 MB | < 200 MB |
| `as` | Assamese | Meta MMS VITS | `mms_tts_asm` | 138 MB | < 200 MB |
| `gu` | Gujarati | Meta MMS VITS | `mms_tts_guj` | 138 MB | < 200 MB |
| `kn` | Kannada | Meta MMS VITS | `mms_tts_kn` | 138 MB | < 200 MB |
| `or` | Odia | Meta MMS VITS | `mms_tts_or` | 138 MB | < 200 MB |
| `doi` | Dogri | Meta MMS VITS | `mms_tts_doi` | 138 MB | < 200 MB |
| `mai` | Maithili | Meta MMS VITS | `mms_tts_mai` | 138 MB | < 200 MB |
| `brx` | Bodo | Indic-TTS Dict | `voice_tts_brx` | < 1 MB | < 50 MB |
| `kok` | Konkani | Indic-TTS Dict | `voice_tts_kok` | < 1 MB | < 50 MB |
| `mni` | Manipuri | Indic-TTS Dict | `voice_tts_mni` | < 1 MB | < 50 MB |
| `sa` | Sanskrit | Indic-TTS Dict | `voice_tts_sa` | < 1 MB | < 50 MB |
| `sat` | Santali | Indic-TTS Dict | `voice_tts_sat` | < 1 MB | < 50 MB |
| `sd` | Sindhi | Indic-TTS Dict | `voice_tts_sd` | < 1 MB | < 50 MB |

---

## ⚖️ Open Source Licensing & Credits
- **Piper Neural Voices**: [Piper](https://github.com/rhasspy/piper) by Michael Hansen (Rhasspy) and contributors. Licenses: **MIT / CC BY 4.0 / Public Domain** (model dataset dependent).
- **Meta MMS VITS Models**: [Meta Massively Multilingual Speech](https://github.com/facebookresearch/fairseq/tree/main/examples/mms) by **Meta AI Research**. License: **CC-BY-NC 4.0** (Creative Commons Attribution-NonCommercial 4.0).
- **Indic Voice Phonetic Dictionaries**: [AI4Bharat Indic-TTS](https://github.com/AI4Bharat/Indic-TTS) by **AI4Bharat (IIT Madras)**. License: **MIT License**.
- **eSpeak-NG**: [eSpeak-NG](https://github.com/espeak-ng/espeak-ng) multi-lingual speech synthesizer. License: **GPL-3.0**.