Text-to-Speech
PyTorch
moss_tts_nano
custom_code

urdu language tts

#1
by ttg73ml - opened

Dear OpenMOSS Team,
​First of all, huge congratulations on the amazing release of MOSS-TTS-Nano. The fact that this model runs so smoothly on a CPU with zero-shot voice cloning capabilities is truly groundbreaking!
​I am writing to earnestly request you to please consider adding support for the Urdu and Hindi languages in your upcoming updates.
​Unfortunately, while there are many TTS models out there, no one has built a truly high-quality, accurate, and native-sounding voice cloning/TTS model for the Urdu language yet. Urdu is the national language of Pakistan and, alongside Hindi, is spoken and understood by hundreds of millions of people globally.
​If you could include Urdu support in a lightweight, high-performance model like MOSS-TTS-Nano, it would be a massive breakthrough and a huge gift for our entire community. We desperately need a high-quality model like yours for our language.
​Please consider this request. Thank you for your incredible contribution to the open-source AI community!
​Best regards,
A passionate user from Pakistan

Our team, NghiStudio, can do that but do you know or have any datasets for Hindi and Urdu with good quality available now?

​Dear Team,
​Thank you so much for your reply! I have done some research and found exactly what you need. Here are some of the best datasets available right now for Urdu and Hindi:

For Urdu:

https://huggingface.co/datasets/humairawan/Urdu-LjSpeech/tree/main

https://huggingface.co/datasets/humairawan/Urdu-NSW/tree/main
https://huggingface.co/datasets/humairawan/UrduSpeech/tree/main/Urdu
https://huggingface.co/datasets/humair025/Urdu-ONYX-WAV/tree/main
https://huggingface.co/datasets/ai4bharat/indicvoices_r/tree/main/Urdu

​For Hindi:
https://huggingface.co/datasets/ai4bharat/indicvoices_r/tree/main/Hindi

​I want to make a very special and earnest request: Please prioritize and build a strong model for the Urdu language. I live in Pakistan, and Urdu is my native language. To be honest, while there are a few Urdu TTS models in the world, they either sound very robotic, or the good ones require heavy GPUs to run.
​The magic of your MOSS-TTS-Nano is that it runs flawlessly on a CPU. That is exactly what our community needs! I highly request you to make a mixed model that seamlessly supports English, Urdu, and Hindi together.
​I am currently building my own AI Web UI under my company name, NZG (nzg73), and I really want to integrate your CPU-friendly model into my release.
​I will be honest: I am not a technical genius or a dataset expert. However, since Urdu is my native language and I know it perfectly, I can offer my personal help. If you need more clean audio data for training or testing, I am more than happy to record my own voice and provide you with high-quality audio samples. Just let me know!
​Lastly, I have one more dream request: Please consider releasing a unified CPU-friendly model in the future that does both Text-to-Speech (TTS) and Speech-to-Text (STT) together. A unified CPU model for this doesn't exist yet!
​Please make the Urdu model powerful. This is my sincere request to you, please!
​Best regards,
​NZG (nzg73)
Email: nzgnzg73@gmail.com

plzz 😣

Sign up or log in to comment