BEIJING, Aug. 25 (Xinhua) -- Chinese researchers have unveiled an open-source text-to-speech artificial intelligence (AI) model that tackles the long-standing challenge of Arabic speech synthesis. The system covers more than 20 regional variants of the language under a unified technical framework.
The model, named Habibi, meaning "my dear friend" in Arabic, was developed by a team of Chinese research institutions, including Shanghai Jiao Tong University, and video-sharing platform Kuaishou.
Spoken by over 400 million people, Arabic has more than 30 regional variants with sharp pronunciation and vocabulary differences. Modern Standard Arabic is used almost exclusively as the official and literary form, while daily conversation relies on dialects from Egypt, Saudi Arabia, Morocco and elsewhere, making unified modeling difficult.
Previous research focused either solely on Modern Standard Arabic or on single dialects. Before Habibi, no open-source text-to-speech system unified these dialects.
For Arabic speakers, dialect voice matters in daily life, said Mahitab Elbakry, an Egyptian news editor in her thirties.
"If my mother wants to use a chatbot to recommend recipes, it is natural for her to interact with it in her dialect," she said. Elbakry also cited online health inquiries, noting that using a dialect would be more accurate and reduce the risk of miscommunication.
"Services tailored to Arabic dialects could become increasingly convenient and widely adopted," she added.
But she noted that while AI products like ChatGPT offer interactive and generative services in Egyptian Arabic, high subscription fees limit access for many users.
The Habibi team has made everything publicly available, including all the model files, the code for training and running it, and the test data, making it "the first such release for a multi-dialect Arabic text-to-speech system," the team said in a paper posted on arXiv, an open-access preprint platform for academic research.
The researchers also claim the AI model can replicate a voice from a brief recording, with no advance training required.
Habibi is the latest example of deepening China-Arab AI collaboration. Chinese tech firms and institutions, including iFLYTEK and Huawei, have been working with Arab partners to advance speech recognition, develop AI models, and optimize dialect synthesis.
Open source has been a key feature of China's AI development. China has proposed the Global AI Governance Initiative and implemented the AI Capacity-Building Action Plan for Good, aiming to lower barriers to global AI adoption and bridge the intelligence divide.
The approach has drawn international attention. Recently, several U.S. tech firms and institutions, including Microsoft, Nvidia and Hugging Face, issued a joint statement supporting open-weight AI models. Microsoft's AI Diffusion Report 2025 also noted that China's "DeepSeek's rise shows that global AI adoption is shaped as much by access and availability as by model quality."
Although the AI Habibi has not yet entered the market, the researchers said it has already surpassed a leading commercial model on major dialect tests, highlighting the competitiveness of an open-source solution.
"We have always been committed to supporting open source," said Chen Xie, a member of the Habibi team and an associate professor at Shanghai Jiao Tong University. "We believe in the power of openness and sharing." ■
