Thomcles/Chatterbox-TTS-Persian-Farsi

Chatterbox Persian-Farsi

training High quality TTS with low ressource data

Chatterbox-TTS-Persian-Farsi is a TTS trained on data that I cleaned, denoised, and filtered.

The total cost of the TTS is $150 on my cloud hardware.

If you find this model useful and high-quality, and would like to support my work, you can send me money via ko-fi, or like it on huggingface.

Dataset : Thomcles/Persian-Farsi-Speech

---

---

demo audios:

"سلام! به آزمایش تبدیل متن به گفتار خوش آمدید."

"سه سیب سرخ روی سینی سیمی است"

"دیروز در تهران باران شد، امروز آفتابی است"

"قیمت لپ‌تاپ جدید من پنجاه میلیون تومان است."

"علی، نرگس و یوسف به دانشگاه شیراز رفتند."

"لطفاً جملهٔ قبل را دوباره تکرار کن، دوباره تکرار کن، دوباره تکرار کن!"

💻 Inference Code

First, download the file from huggingface and place it in the current directory.

from chatterbox import mtl_tts
import torchaudio as ta
from safetensors.torch import load_file as load_safetensors

device = "cpu" # or mps or cuda

multilingual_model = mtl_tts.ChatterboxMultilingualTTS.from_pretrained(device=device)

# ----
# Then download the file from huggingface and place it in the current directory.
# ----


t3_state = load_safetensors("t3_fa_v2.safetensors", device="cpu")
multilingual_model.t3.load_state_dict(t3_state)
multilingual_model.t3.to(device).eval()

persian_text = "سلام! به آزمایش تبدیل متن به گفتار خوش آمدید."
wav_persian = multilingual_model.generate(persian_text, language_id=None)
ta.save("test-fa.wav", wav_persian, model.sr)

Acknowledgements

Thanks to @phamed for the Persian training data.

contact :

e-mail : [email protected]

☕ Support

I trained this model from my own financial resources with the sole aim of offering it to the huggingface open source community.

This model has cost me a lot of money. If you find this checkpoint useful and would like to support my work, you can do it via Ko-fi: