Nidum-Madurai-Tamil-TTS

🔊 Tamil Text-to-Speech (TTS) Model by Nidum
🧪 Live Demo: Try it on Hugging Face Spaces

🗣️ Overview

This is a high-quality Tamil Text-to-Speech (TTS) model developed by Nidum. It generates clear, natural-sounding Tamil speech from input text and is suitable for voice assistants, screen readers, language learning apps, and content narration.

---

🚀 Features

• ✅ Converts Tamil script text to speech
• ✅ Natural and expressive voice
• ✅ Option to choose male or female voice
• ✅ Easy-to-use demo via Hugging Face Spaces

---

🧪 Live Demo

👉 Click here to try the demo

Type Tamil text, select speaker, click Generate, and listen instantly!

---

🧑‍🎤 Speakers

Speaker ID	Voice
0 speaker	Male
1 speaker	Female

Use the appropriate speaker ID in your prompt like this:

0 speaker: வணக்கம்!

---

💻 Usage (Code Example)

import torch
import soundfile as sf
from transformers import AutoModelForCausalLM, AutoTokenizer
from snac import SNAC

# Path to your fine-tuned model
fine_tuned_checkpoint = "<Model_ID>"

# Load model & tokenizer
print("Loading model...")
model = AutoModelForCausalLM.from_pretrained(fine_tuned_checkpoint, torch_dtype=torch.bfloat16).cuda()
tokenizer = AutoTokenizer.from_pretrained(fine_tuned_checkpoint)

# Load SNAC vocoder
print("Loading SNAC model...")
snac_model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz").to("cpu")

# Input prompt with speaker
prompts = [
    "0 speaker: வணக்கம்! இந்த பயன்பாட்டை பயன்படுத்தி உங்கள் உரையை குரலாக்கலாம்."
]

# Tokenize
all_input_ids = [tokenizer(p, return_tensors="pt").input_ids for p in prompts]
start_token = torch.tensor([[128259]], dtype=torch.int64)
end_tokens = torch.tensor([[128009, 128260]], dtype=torch.int64)
all_modified_input_ids = [torch.cat([start_token, ids, end_tokens], dim=1) for ids in all_input_ids]

# Pad
max_length = max([ids.shape[1] for ids in all_modified_input_ids])
all_padded_tensors, all_attention_masks = [], []
for modified_input_ids in all_modified_input_ids:
    padding = max_length - modified_input_ids.shape[1]
    padded_tensor = torch.cat([torch.full((1, padding), 128263, dtype=torch.int64), modified_input_ids], dim=1)
    attention_mask = torch.cat([torch.zeros((1, padding), dtype=torch.int64), torch.ones((1, modified_input_ids.shape[1]), dtype=torch.int64)], dim=1)
    all_padded_tensors.append(padded_tensor)
    all_attention_masks.append(attention_mask)

input_ids = torch.cat(all_padded_tensors, dim=0).cuda()
attention_mask = torch.cat(all_attention_masks, dim=0).cuda()

# Generate
print("Generating speech...")
with torch.no_grad():
    generated_ids = model.generate(
        input_ids=input_ids,
        attention_mask=attention_mask,
        max_new_tokens=4800,
        do_sample=True,
        temperature=1,
        top_p=1,
        repetition_penalty=1.1,
        num_return_sequences=1,
        eos_token_id=128258,
    )

# Extract audio codes
print("Parsing output...")
token_indices = (generated_ids == 128257).nonzero(as_tuple=True)
if len(token_indices[1]) > 0:
    last_occurrence_idx = token_indices[1][-1].item()
    cropped_tensor = generated_ids[:, last_occurrence_idx + 1:]
else:
    cropped_tensor = generated_ids

processed_rows = [row[row != 128258] for row in cropped_tensor]

code_lists = []
for row in processed_rows:
    row_length = row.size(0)
    new_length = (row_length // 7) * 7
    trimmed_row = row[:new_length] - 128266
    code_lists.append(trimmed_row.tolist())

# Decode with SNAC
def redistribute_codes(code_list):
    layer_1, layer_2, layer_3 = [], [], []
    for i in range((len(code_list) + 1) // 7):
        layer_1.append(code_list[7 * i])
        layer_2.append(code_list[7 * i + 1] - 4096)
        layer_3.append(code_list[7 * i + 2] - (2 * 4096))
        layer_3.append(code_list[7 * i + 3] - (3 * 4096))
        layer_2.append(code_list[7 * i + 4] - (4 * 4096))
        layer_3.append(code_list[7 * i + 5] - (5 * 4096))
        layer_3.append(code_list[7 * i + 6] - (6 * 4096))
    codes = [
        torch.tensor(layer_1).unsqueeze(0),
        torch.tensor(layer_2).unsqueeze(0),
        torch.tensor(layer_3).unsqueeze(0)
    ]
    return snac_model.decode(codes)

print("Decoding speech...")
audio_samples = [redistribute_codes(codes) for codes in code_lists]

# Save audio
for i, samples in enumerate(audio_samples):
    audio_data = samples.detach().squeeze().to("cpu").numpy()
    sf.write(f"output_{i}.wav", audio_data, samplerate=24000)
    print(f"Audio {i} saved as output_{i}.wav")

print("Done!")

---

📬 Contact

For questions, feedback, or collaboration:
📧 [email protected]