Indic-TTS-v1

Indic-TTS-v1 is a multilingual Text-to-Speech (TTS) model trained on dataset covering 11 major Indic languages. Built upon the Qwen3-TTS architecture, it generates natural-sounding speech from input text across various South Asian languages and dialects.

---

Model Details

• Model Type: Text-to-Speech (TTS)
• Base Architecture: Qwen3-TTS
• Languages Supported: 11 Indic Languages (Hindi, Bengali, Marathi, Telugu, Tamil, Gujarati, Kannada, Malayalam, Odia, Punjabi, Urdu)
• License: Apache 2.0

---

Supported Languages

The model has been trained on speech data from the following 11 Indic languages:

Language	ISO Code	Language Family
Hindi	hi	Indo-Aryan
Bengali	bn	Indo-Aryan
Marathi	mr	Indo-Aryan
Telugu	te	Dravidian
Tamil	ta	Dravidian
Gujarati	gu	Indo-Aryan
Kannada	kn	Dravidian
Malayalam	ml	Dravidian
Odia	or	Indo-Aryan
Punjabi	pa	Indo-Aryan
Urdu	ur	Indo-Aryan

---

Usage

Quickstart

Below is a basic example of how to load and use the model for audio synthesis:

import torch
# Example model loading (adjust import based on your specific implementation/pipeline)
# from transformers import AutoModelForTextToWaveform, AutoTokenizer

model_id = "2023310197mehak/indic-tts-v1"

# Load model & tokenizer
# tokenizer = AutoTokenizer.from_pretrained(model_id)
# model = AutoModelForTextToWaveform.from_pretrained(model_id)

# Synthesize speech
# text = "नमस्ते, आप कैसे हैं?"  # Example Hindi text
# inputs = tokenizer(text, return_tensors="pt")
# with torch.no_grad():
#     audio = model.generate(**inputs)