Speech synthesis overview

Model Studio converts text to speech using system voices or custom voices created through voice cloning or voice design. You can also control synthesis effects such as speed, emotion, and pronunciation.

Model selection

Choose a model based on whether your application needs real-time interaction, content production, or fine-grained reading controls.

Requirement	Recommended model	Key capabilities
Voice assistants, real-time conversations, and customer service	`qwen-audio-3.1-tts-flash`	Designed for low-latency interaction. Supports system voices, voice cloning, voice design, natural-language instructions, and emotion and paralinguistic tags.
Audiobooks, dubbing, brand voices, and content production	`qwen-audio-3.1-tts-plus`	Focuses on audio quality, naturalness, and expressive detail. Supports system voices, voice cloning, voice design, and expressive controls.

For the model IDs in each family, see Supported models.

Voice selection

Visit the Voice Plaza to listen to voices. For system voices available to each model, voice values, and audio samples, see the voice lists in the following table.

Voice source	Preparation	Use cases	Documentation
System voice	Select an existing voice. No audio sample or voice description is required.	Customer service, audiobooks, and broadcasts for which an existing voice meets your needs.	Qwen-Audio-TTS voices CosyVoice voices Qwen-TTS voices
Cloned voice	Provide an audio sample of the target speaker.	Reuse a spokesperson or presenter’s voice, or create personalized assistants and virtual presenters.	Voice cloning
Designed voice	Describe the desired voice, such as “a warm, low-pitched female voice.”	Create a brand or character voice for a game or animation when no recording is available.	Voice design

Voice design creates a reusable voice. To adjust the speed, emotion, or speaking style for a single synthesis request, use instruction control.

Synthesis modes

Synthesis modes differ in how text is submitted and audio is returned: text can be sent in one complete request or in successive segments, and audio can be returned all at once or as it is generated. Streaming audio output and streaming text input are separate capabilities.

Mode	Text input	Audio output	Use cases
Non-streaming	Send the complete text once.	Retrieve the complete audio after synthesis finishes.	Content production without playback during synthesis.
Unidirectional streaming	Send the complete text once.	Receive audio in chunks during synthesis.	Text is ready and playback should start as early as possible.
Bidirectional streaming	Send text segments as they become available.	Receive audio in chunks during synthesis.	Real-time interaction with incrementally generated text, such as LLM output.

Non-real-time synthesis: Submit the complete text in one request and receive the audio as a complete result or a stream. See Non-real-time speech synthesis.
Real-time synthesis: Submit text in successive segments, continuously, or all at once. See Real-time speech synthesis. For mobile and other real-time client applications, use the Realtime API overview to select an available access protocol. AOQ (AI over QUIC) is optimized for low latency and weak networks, with built-in echo cancellation and noise suppression.

Synthesis controls

In addition to selecting a voice, use parameters, instructions, or tags to adjust speech output.

Desired effect	Control method
Speed, pitch, and volume	Set the corresponding synthesis parameters provided by the model.
Emotion, style, and delivery	Use natural-language instructions, such as “speak gently at a slower pace.”
Local emotion and expressions such as laughter and breathing	Add emotion and paralinguistic tags to the text.
Pronunciation, pauses, and number reading	Use SSML tags for fine-grained reading controls.
Mathematical formula reading	Express formulas in LaTeX.

For parameters, support requirements, and examples, select the guide for your synthesis mode:

Supported models

Select the region to view the model IDs in each family.

Beijing

Qwen-Audio-TTS: qwen-audio-3.1-tts-plus, qwen-audio-3.1-tts-flash, qwen-audio-3.0-tts-plus, qwen-audio-3.0-tts-flash
CosyVoice: cosyvoice-v3.5-plus, cosyvoice-v3.5-flash, cosyvoice-v3-plus, cosyvoice-v3-flash, cosyvoice-v2
Qwen3-TTS: qwen3-tts-flash-realtime, qwen3-tts-flash
MiniMax: MiniMax/speech-2.8-hd, MiniMax/speech-2.8-turbo, MiniMax/speech-02-hd, MiniMax/speech-02-turbo

Singapore

Qwen-Audio-TTS: qwen-audio-3.1-tts-plus, qwen-audio-3.1-tts-flash, qwen-audio-3.0-tts-plus, qwen-audio-3.0-tts-flash
CosyVoice: cosyvoice-v3-plus, cosyvoice-v3-flash
Qwen3-TTS: qwen3-tts-flash-realtime, qwen3-tts-flash

Languages and dialects

Choose a voice that supports the language or dialect you want to synthesize. The following tables group support by model family and voice type.

Qwen-Audio-TTS

Voice type Speech output languages and dialects

Voice type	Speech output languages and dialects
System voice	Support languages such as Chinese and English. Availability varies by model and voice. See the Qwen-Audio-TTS voice list.
Cloned voice	Chinese (Mandarin, Cantonese, Chongqing, Northeastern, Gansu, Guizhou, Zhejiang, Hebei, Henan, Hubei, Hunan, Jiangxi, Ningbo, Ningxia, Qingdao, Shaanxi, Shanxi, Shandong, Shanghainese, Sichuan, Yunnan), English, Japanese, Korean, Russian, French, German, Portuguese, Thai, Indonesian, Vietnamese, Spanish, Italian, Malaysian, Filipino, and Arabic (dialects are configured through instruction control) Other languages (supported by `qwen-audio-3.1-tts-flash`): Bulgarian, Dutch, Hungarian, Romanian, Danish, Croatian, Greek, Czech, Norwegian, Swedish, Hindi, Slovak, Finnish, Urdu, Serbian, Mexican Spanish, Lithuanian, Latvian, Turkish, Persian, Hebrew, Mongolian, Slovenian, Azerbaijani, Polish, Catalan, Khmer, Brazilian Portuguese, Lao, Georgian, Afrikaans, Nepali, Belarusian, Malayalam, Luxembourgish, Albanian, Tamil, Sinhala, Telugu, Estonian, Cebuano, Bosnian, Burmese, Armenian, Latin, Icelandic, Maltese, Swahili, Kazakh, Seychellois Creole, Ukrainian.
Designed voice	Chinese (Mandarin), English

System voice

Support languages such as Chinese and English. Availability varies by model and voice. See the Qwen-Audio-TTS voice list.

Cloned voice

Chinese (Mandarin, Cantonese, Chongqing, Northeastern, Gansu, Guizhou, Zhejiang, Hebei, Henan, Hubei, Hunan, Jiangxi, Ningbo, Ningxia, Qingdao, Shaanxi, Shanxi, Shandong, Shanghainese, Sichuan, Yunnan), English, Japanese, Korean, Russian, French, German, Portuguese, Thai, Indonesian, Vietnamese, Spanish, Italian, Malaysian, Filipino, and Arabic (dialects are configured through instruction control)

Other languages (supported by qwen-audio-3.1-tts-flash): Bulgarian, Dutch, Hungarian, Romanian, Danish, Croatian, Greek, Czech, Norwegian, Swedish, Hindi, Slovak, Finnish, Urdu, Serbian, Mexican Spanish, Lithuanian, Latvian, Turkish, Persian, Hebrew, Mongolian, Slovenian, Azerbaijani, Polish, Catalan, Khmer, Brazilian Portuguese, Lao, Georgian, Afrikaans, Nepali, Belarusian, Malayalam, Luxembourgish, Albanian, Tamil, Sinhala, Telugu, Estonian, Cebuano, Bosnian, Burmese, Armenian, Latin, Icelandic, Maltese, Swahili, Kazakh, Seychellois Creole, Ukrainian.

Designed voice

Chinese (Mandarin), English

CosyVoice

Voice type	Speech output languages and dialects
System voice	`cosyvoice-v3.5-plus` / `cosyvoice-v3.5-flash`: Not supported `cosyvoice-v3-plus`(varies by voice): Chinese (Mandarin), English `cosyvoice-v3-flash`(varies by voice; some dialects are directly supported by system voices, others through instruction control): Chinese (Mandarin, Cantonese, Northeastern, Henan, Hunan, Shaanxi, Shandong, Sichuan, Anhui, Hokkien), English , Japanese, Korean, and Indonesian `cosyvoice-v2`(varies by voice): Chinese (Mandarin, Cantonese, Northeastern, Hokkien, Shaanxi), English, Japanese, and Korean NoteFor the languages, dialects, and audio samples of individual system voices, see the CosyVoice voice list.
Cloned voice	`cosyvoice-v3.5-plus` / `cosyvoice-v3.5-flash`(dialects are configured through instruction control): Chinese (Mandarin, Cantonese, Northeastern, Gansu, Guizhou, Henan, Hubei, Jiangxi, Hokkien, Ningxia, Shanxi, Shaanxi, Shandong, Shanghainese, Sichuan, Tianjin, Yunnan), English, French, German, Japanese, Korean, Russian, Portuguese, Thai, Indonesian, and Vietnamese `cosyvoice-v3-plus`: Chinese (Mandarin, Cantonese, Northeastern, Gansu, Guizhou, Henan, Hubei, Jiangxi, Hokkien, Ningxia, Shanxi, Shaanxi, Shandong, Shanghainese, Sichuan, Tianjin, Yunnan), English, French, German, Japanese, Korean, and Russian `cosyvoice-v3-flash`(dialects are configured through instruction control): Chinese (Mandarin, Cantonese, Northeastern, Gansu, Guizhou, Henan, Hubei, Jiangxi, Hokkien, Ningxia, Shanxi, Shaanxi, Shandong, Shanghainese, Sichuan, Tianjin, Yunnan), English, French, German, Japanese, Korean, Russian, Portuguese, Thai, Indonesian, and Vietnamese `cosyvoice-v2`: Chinese (Mandarin), English
Designed voice	`cosyvoice-v3.5-plus` / `cosyvoice-v3.5-flash`: Chinese (Mandarin), English `cosyvoice-v3-plus`: Chinese (Mandarin), English `cosyvoice-v3-flash`: Chinese (Mandarin), English `cosyvoice-v2`: Not supported

Qwen3-TTS

Voice type	Speech output languages and dialects
System voice	Chinese (Mandarin, Beijing, Shanghainese, Sichuan, Nanjing, Shaanxi, Hokkien, Tianjin, Cantonese — varies by voice), English, German, Italian, Portuguese, Spanish, Japanese, Korean, French, and Russian NoteFor the languages, dialects, and audio samples of individual system voices, see the Qwen-TTS voice list.
Cloned voice	Not supported
Designed voice	Not supported