Chinese AI Habibi masters Moroccan Darija and challenges ElevenLabs
Chinese researchers have launched “Habibi,” the first open-source AI model capable of synthesizing more than 20 Arabic dialects, including Moroccan Darija. This innovation from Shanghai Jiao Tong University outperforms current commercial benchmarks.
Developed by the X-LANCE laboratory, this speech synthesis system uses a “zero-shot” capability to clone a voice from a simple audio clip. According to the study’s results, Habibi already outperforms solutions from American company ElevenLabs in generating Darija and other major regional dialects.
On Bladi.net : Arabic Reigns Supreme: World’s Richest Language Boasts 12 Million Words
The project tackles the complexity of spoken Arabic, which is often very different from the official language used in education. “The ultimate bottleneck comes down to data,” explains Chen Yushen, the lead researcher, highlighting the scarcity of high-quality recordings capturing local linguistic specificities.
This breakthrough targets the massive social media and e-commerce market in the Middle East. The model could transform localized content production for TikTok and podcasts, thereby strengthening the technological influence and image of Chinese AI throughout the MENA region.
On Bladi.net : Love is Blind, Habibi: an Iraqi candidate insults Moroccan women
Although pioneering, the team considers this work a foundation to be refined through academic research. “We hope that by making the model open source, we can gather broader feedback from the community,” Chen said, encouraging collaboration with native speakers.
