Alibaba's Qwen team has released a significant update to its text-to-speech system, Qwen3-TTS, expanding its repertoire of voices, languages, and dialects. The update, named Qwen3-TTS-Flash, aims to provide more natural and expressive synthetic speech for developers and businesses through its API.

The model now offers a selection of 49 distinct voice "timbres," designed to cover a range of genders, ages, and character profiles—from a playful, quirky persona to a wise elder or a strict instructor. This move addresses a growing demand for more varied and character-driven synthetic voices in applications like audiobooks, virtual assistants, and interactive media.

Key Enhancements in the Update:

· Broad Language and Dialect Support: Beyond supporting ten major languages including Chinese, English, Japanese, and Spanish, the model now incorporates nine Chinese dialects. This includes regional accents like Cantonese, Sichuanese, and the Beijing dialect, aiming to authentically capture local linguistic nuances.

· Improved Speech Naturalness: The team reports substantial improvements in the system's ability to adapt speech rate and prosody based on text input, closing the gap between synthetic and human speech.

· Competitive Performance Claims: Alibaba states that on a standard multilingual test set, Qwen3-TTS achieves a lower average word error rate than competing services from MiniMax, ElevenLabs, and OpenAI's GPT-4o Audio Preview.

The update is available via the Qwen API, with sample code provided for developers to integrate the service. By significantly expanding vocal diversity and dialect support, Alibaba is positioning its TTS technology for a wider global and regional market, catering to more personalized and locally relevant audio content creation.