Qwen3-TTS-Flash: Multi-timbre & Multi-lingual & Multi-dialect Speech Synthesis.

DASHSCOPE

HUGGING FACE DEMO

MODELSCOPE DEMO

Qwen3-TTS-Flash is a flagship text-to-speech model that supports multi-timbre, multi-lingual, and multi-dialect speech synthesis. It aims to produce natural and expressive speech and is available via Qwen API.

Key Features:

• Excellent Chinese-English Stability: Qwen3-TTS-Flash achieves SOTA performance in Chinese and English stability on seed-tts-eval test set, surpassing SeedTTS, MiniMax, and GPT-4o-Audio-Preview.

• Superior Multilingual Performance: In terms of multilingual stability and timbre similarity, Qwen3-TTS-Flash achieves SOTA WER for Chinese, English, Italian, and French on MiniMax TTS multilingual test set, significantly lower than MiniMax, ElevenLabs, and GPT-4o-Audio-Preview. Speaker similarity in English, Italian, and French also significantly surpasses MiniMax, ElevenLabs, and GPT-4o-Audio-Preview.

• High Expressiveness: Qwen3-TTS-Flash delivers highly expressive and human-like voice timbres, providing stable and reliable speech output that faithfully follows the input text.

• Rich Timbres and Languages: Qwen3-TTS-Flash offers 17 distinct speaker options, each supporting 10 languages.

• Multi-dialect Support: Qwen3-TTS-Flash supports dialect generation, including Mandarin, Hokkien, Wu dialect, Cantonese, Sichuanese, Beijing dialect, Nanjing dialect, Tianjin dialect, and Shaanxi dialect.

• Tone Adaptation: Trained on large-scale datasets, Qwen3-TTS-Flash can automatically adjust tone to match the input text.

• Strong Robust: Qwen3-TTS-Flash automatically processes complex texts by extracting key information, demonstrating robustness to complex and diverse formats.

• Fast Generation: Qwen3-TTS-Flash has very low first-packet latency with single-threaded first-packet latency as low as 97ms.

Samples#

Here are some multilingual samples by a single speaker:

Here are some samples by Chinese and English speakers:

Here are some samples by dialect speakers:

Here are some mixed multilingual samples:

Language Support#

Qwen3-TTS-Flash enables multilingual speech generation for multi speakers through a unified model:

• Chinese: Mandarin, Hokkien, Wu dialect, Cantonese, Sichuanese, Beijing dialect, Nanjing dialect, Tianjin dialect, and Shaanxi dialect.

• English: British accent, American accent, and various regional English accents.

• Other Languages: French, German, Russian, Italian, Spanish, Portuguese, Japanese, and Korean.

Performance#

We conducted a comprehensive evaluation of Qwen3-TTS-Flash in terms of speech stability and timbre similarity, and it achieves SOTA performance across many results. Specifically, on the seed-tts-eval test set, Qwen3-TTS-Flash demonstrates SOTA stability for both Chinese and English, outperforming SeedTTS, MiniMax, and GPT-4o-Audio-Preview. Furthermore, on MiniMax TTS multilingual test set, Qwen3-TTS-Flash achieves SOTA WER for Chinese, English, Italian, and French, significantly lower than those of MiniMax, ElevenLabs, and GPT-4o-Audio-Preview. In terms of speaker similarity for English, Italian, and French, Qwen3-TTS-Flash also surpasses the aforementioned models, demonstrating remarkable performance in both multi-language stability and timbre similarity.

We introduce several architectural upgrades and acceleration strategies to achieve lower first-packet latency and faster generation speed.

How to use#

Use Qwen3-TTS-Flash with Qwen API is simple. We demonstrate a code snippet for you to play with it below:}}

What's Next#

Qwen3-TTS-Flash revolutionizes the speech synthesis experience, achieving industry-leading performance in multi-speaker capabilities, multilingual support, multi-dialect adaptation, and text processing robustness. We look forward to seeing it empower more innovative applications. Everyone is invited to experience it and share valuable feedback!