atx-labs/Vyom-TTS-Hindi-3B

Vyom TTS HI

Vyom TTS HI is a high-quality Text-to-Speech (TTS) model tailored for the Hindi language, featuring an expressive female voice. Built by fine-tuning the 3B speech-LLM base model (unsloth/orpheus-3b-0.1-ft) in 16-bit precision, it produces 24 kHz high-fidelity regional speech synthesis via the SNAC neural vocoder.

Model Details

Model Description

• Developed by: OpenXfi by ATX Labs
• Model Type: Fine-tuned Speech-LLM (Base: unsloth/orpheus-3b-0.1-ft)
• Fine-Tuning Method: 16-Bit LoRA (r=64, lora_alpha=128, rsLoRA)
• Language(s): Hindi (hi)
• Voice Characteristics: Female voice
• Audio Output: 24 kHz WAV (SNAC 24kHz Vocoder)
• License: MIT

Uses

Direct Use

• Generating localized Hindi voiceovers for regional content, audiobooks, and media.
• Building conversational AI agents and regional virtual assistants.
• Accessibility tools and reading assistants for Hindi text.

Out-of-Scope Use

• Voice Cloning: The model is not designed or authorized for unauthorized voice cloning or deepfake generation.
• Malicious Imitation: Generation of deceptive audio or unauthorized impersonation of individuals.

Evaluation & Benchmarks

Evaluation Dataset

The model was evaluated on a benchmark dataset of 210 Hindi evaluation prompts categorized into three core domains:

1. Hindi Complex (60 prompts): Multi-clause sentences, high-level vocabulary, and long-form text.
2. Hindi Normal (100 prompts): General everyday conversational Hindi text.
3. Hindi Numbers (50 prompts): Sentences containing numbers written as digits and spelled-out words.

Evaluation Methodology

All synthesized 24 kHz Hindi audio samples were transcribed using AI4Bharat IndicConformer Multilingual ASR (ai4bharat/indic-conformer-600m-multilingual) to measure Word Error Rate (WER) and Character Error Rate (CER).

Empirical Results

Category	Prompt Count	Word Error Rate (WER)	Character Error Rate (CER)
Hindi Complex	60	0.1657	0.0401
Hindi Normal	100	0.2913	0.0579
Hindi Numbers	50	0.4589	0.2885
Overall Average	210	0.2953	0.1077

Bias, Risks, and Limitations

• Numerical Formatting: Basic numerical functionality provided. Optimal synthesis performance is achieved when numbers are written out as text words in Hindi (e.g., "एक सौ पचास") rather than raw numeric digits (e.g., "१५०").
• English Loanwords & Acronyms: Code-mixed English words or uncommon acronyms may produce minor phonetic artifacts.

How to Get Started with the Model

Prerequisites

Install the required dependencies:

pip install torch transformers snac soundfile

Path A: Direct Model Generation (generate_speech)

import os, torch, soundfile as sf
from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "atx-labs/voice-hindi-female"
hf_token = os.environ.get("HF_TOKEN")

tokenizer = AutoTokenizer.from_pretrained(repo_id, token=hf_token)
model = AutoModelForCausalLM.from_pretrained(
    repo_id, trust_remote_code=True, torch_dtype=torch.bfloat16, token=hf_token
).to("cuda").eval()

# Synthesize speech directly
waveform = model.generate_speech("नमस्ते, आपका स्वागत है।", tokenizer)

# Save 24 kHz WAV audio
sf.write("output_hi.wav", waveform, 24000)
print("✅ Audio generated and saved to output_hi.wav!")

Path B: transformers Pipeline

import os, torch, soundfile as sf
from transformers import pipeline

repo_id = "atx-labs/voice-hindi-female"
hf_token = os.environ.get("HF_TOKEN")

# Initialize TTS pipeline
tts = pipeline("text-to-speech", model=repo_id, trust_remote_code=True, device=0, token=hf_token)

# Synthesize speech
output = tts("नमस्ते, आपका स्वागत है।")

# Save 24 kHz WAV audio
sf.write("output_hi.wav", output["audio"], output["sampling_rate"])
print("✅ Audio generated and saved to output_hi.wav!")

Contributors

• Mrugaja Joshi — @MrugajaJ
• Guruprasad Pathak — @grpathak22

About OpenXFI

Vyom TTS is released as part of OpenXFI, an open-source Indic AI initiative by ATX Labs. OpenXFI builds and openly releases foundational models, benchmarks, and datasets for Indian languages — with the goal of making high-quality Indic AI accessible to researchers, developers, and institutions across the country.

Explore the wider release, including our Indic evaluation benchmark and open datasets, at openxfi.com.

Citation

If you use this model in your research or applications, please cite both the model and the OpenXFI initiative:

@misc{vyomttsxfi2026,
  title        = {Vyom TTS: Open Indic Text-to-Speech Models},
  author       = {Joshi, Mrugaja and Pathak, Guruprasad},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/atx-labs}},
  note         = {Released as part of OpenXFI by ATX Labs}
}

@misc{openxfi2026,
  title        = {OpenXFI: Shaping India's AI, Together},
  author       = {{ATX Labs}},
  year         = {2026},
  url          = {https://openxfi.com}
}

We also encourage citing the upstream work this model builds on — the Orpheus base model and the SNAC neural audio codec.

Contact

• Website: openxfi.com
• ATX Labs: atxlabs.ai
• Issues & feedback: open a discussion on this model's Community tab