Qwen3-TTS 发布音色设计与克隆模型
阿里 Qwen 团队推出 Qwen3-TTS-VD-Flash(音色设计)和 Qwen3-TTS-VC-Flash(音色克隆)两款新模型。前者支持通过自然语言指令精细调控音色、情感与人设,后者实现 3 秒级多语言音色克隆。基准测试显示其在 InstructTTS-Eval 和多语言 WER 指标上优于 GPT-4o-mini-tts 及 ElevenLabs 等竞品。
报道正文
本文目录9 个章节
Qwen3-TTS-VD-Flash HF DEMO Qwen3-TTS-VD-Flash MODELSCOPE DEMO Qwen3-TTS-VC-Flash HF DEMO Qwen3-TTS-VC-Flash MODELSCOPE DEMO
Qwen3-TTS 家族新推出两款模型,音色创造模型Qwen3-TTS-VD-Flash(可通过Qwen API访问)和音色克隆模型Qwen3-TTS-VC-Flash(可通过Qwen API访问)。
主要特点:
- 音色创造:Qwen3-TTS-VD-Flash 支持复杂自然语言指令输入,实现对音色、韵律、情感、人设等的精细化调控,实现从“说什么”到“如何说”的全面掌控,可以让用户自由的定义想要的音色,彻底摆脱只能进行根据已有的音色进行克隆或者只能选择固定的一部分预设音色。在 InstructTTS-Eval 中综合表现显著优于 GPT-4o-mini-tts、Mimo-audio-7b-instruct,在角色扮演测试中也超越 Gemini-2.5-pro-preview-tts。
- 音色克隆:Qwen3-TTS-VC-Flash 支持3s级别音色克隆,并且可以基于克隆的音色生成中文、英文、德语、意大利语、葡萄牙语、西班牙语、日语、韩语、法语、俄语等10大主流语言。在MiniMax TTS Multilingual Test Set上,平均词错误率(WER)全面优于MiniMax、ElevenLabs及GPT-4o-Audio-Preview。
- 高表现力:Qwen3-TTS-VD-Flash 和 Qwen3-TTS-VC-Flash 具备高表现力的拟人化音色,能够稳定、可靠地输出高度契合输入文本的语音内容,并根据文本语义自动调节语气节奏,呈现自然生动的表达效果。
- 鲁棒的文本能力:Qwen3-TTS-VD-Flash 和 Qwen3-TTS-VC-Flash 具备强大的文本解析能力,可自动处理复杂文本结构,精准提取关键信息,对多样化、非规范化的文本格式展现出较强的鲁棒性。
Qwen3-TTS-VD-Flash#
Qwen3-TTS 支持通过自然语言描述生成定制化的音色形象。用户可以随意输入声学属性、人设描述、背景信息等自由描述,轻松创造出自己期望的声音形象。
指标#
可控生成:在 InstructTTS-Eval 中,Qwen3-TTS 综合表现显著优于 GPT-4o-mini-tts、Mimo-audio-7b-instruct,在角色扮演测试中也超越 Gemini-2.5-pro-preview-tts。

样例#
| 控制类型 | 控制指令 | 合成文本 | 合成样例 |
|---|---|---|---|
| 声学属性:积极/消极 | 模仿电视购物主持人,中年男性,声音洪亮有激情,语速极快,音调夸张上扬,用极具煽动性的语气来介绍产品,营造出紧迫感和抢购氛围。 | 不要九百九十八,也不要八百八十八,今天只要九十八!对,你没有听错,只要九十八!赶快拿起电话订购吧! | |
| Male, middle-aged, booming baritone - hyper-energetic infomercial voice with rapid-fire delivery and exaggerated pitch rises, dripping with salesmanship | Not nine hundred ninety-nine dollars! Not seven hundred ninety-nine! today, it's just $98! That's right, you heard me, ONLY NINETY-EIGHT DOLLARS! Don't wait, don't hesitate—pick up the phone and call NOW! | ||
| 展现出悲苦沙哑的声音质感,语速偏慢,情绪浓烈且带有哭腔,以标准普通话缓慢诉说,情感强烈,语调哀怨高亢,音高起伏大。 | 这些年代受的苦,就跟你说上十天半个月也说不完。 | ||
| Male, 30s, strained tenor - breathy sobs interrupt speech, pitch swings wildly between whispers and wails | The suffering I've endured... I could talk for days, weeks even, and still not scratch the surface. | ||
| 人设扮演:简洁/丰富 | 邪恶女魔头 | 哥哥,你回来啦,人家等了你好久好久了,要抱抱! | |
| Playful Homebody Sis | Big brooo, you're finally home! I've been waiting forever! Gimme a hug, pwease~! | ||
| 角色姓名:陈远山 身份背景:某国家重点科研项目首席顾问,年近七十的资深战略科学家。曾参与国家重大科技攻关工程,历经数十年风雨,见证了从落后追赶到自主创新的艰难历程。现任国家科技咨询委员会终身荣誉委员,仍坚持在一线培养青年人才,为国家战略发展建言献策。 外貌特征:身形挺拔,两鬓斑白,眉宇间刻着岁月沉淀的坚毅。常着深色中山装或简洁正装,眼神沉静而锐利,举手投足间自带威严与从容。 性格特质:意志如钢,信念坚定,面对挑战从不退缩;胸怀家国,心系民族未来,将个人命运与国家兴衰紧密相连;严谨自律,言出必行,话语中充满责任感与历史担当;外冷内热,表面严肃,实则对后辈寄予厚望,甘为人梯。 人生信条:“我们这一代人,不是为了站在光里,而是为了把路铺到光里。” | 有些事,只要国家需要,就得有人扛起来。 我们那一代人,是背着泥土铺路的; 你们要做的,是让这条路,通向星辰大海。 | ||
| role: Mid-level Corporate Project Manager. gender: Male. pitch: Dynamic male pitch, starting mid-high with agitation, transitioning to a lower declarative range, and spiking upwards with intense emphasis such as 'so finished!'. speed: Variable speaking rate; initially rapid during agitated states ('Damn it!'), slowing for declarative statements like 'I'm done', then accelerating again with strong emotional delivery ('so finished!').. volume: Significant dynamic range; initially loud and forceful, briefly softening to a firm conversational level, then escalating to shouting at points of high emphasis like 'so finished!'. age: Middle-aged adult. clarity: Consistently clear articulation, maintained even during rapid or loud emotional expressions.. fluency: Fluent and coherent speech, with pauses and pacing that align with the expressed emotional state.. accent: General American English. texture: Predominantly forceful, becoming strained during agitated outbursts and shouting, otherwise resonant and firm during calmer declarations.. emotion: Starts with pronounced frustration and exasperation ('Damn it!'), shifts to resolute decisiveness ('I'm done'), culminating in an intensely emphatic declaration of finality ('I am so finished!').. tone: Begins as agitated and questioning, transitions to assertive and declarative, and concludes with a highly emphatic and intense quality.. personality: Assertive and emotionally expressive, demonstrating a build-up of frustration leading to a decisive, forceful resolution.. | So am I damn it. I mean, come on. It's just, you know what I'm through with this. I'm done. Finished, I'm out of here. I am so finished. Those were coming when the health. | ||
| 背景信息 | 《少年闰土》是节选自鲁迅1921年写的短篇小说《故乡》中的一段插叙, 主人公以鲁迅的童年伙伴章运水为原型。 《少年闰土》的题目是被选入小学语文教材后编者加的,自1980年起,它基本上一直保留在小学语文教材中, 目前被选入小学语文教材统编版六年级上册。 《少年闰土》以回忆的方式展开, 刻画出一个机敏勇敢、见多识广的闰土形象。 全文按照“记忆—相识—相处—相别”的顺序书写, 依次介绍了“我”记忆中看瓜刺猹的闰土、初次相识时的闰土、给“我”讲新鲜事的闰土, 少年闰土给“我”的童年生活带来了无穷的新奇与乐趣。 作者善用白描手法,多用直接引语, 其中又蕴含丰富的情感,既有“我”在相见之前对闰土的盼望,也有相处过程中对闰土的羡慕和向往,还有分别时的依依不舍。在短短的相处中,“我”和闰土彼此之间结下深厚情谊。 | 深蓝的天空中挂着一轮金黄的圆月,下面是海边的沙地,种着一望无际的碧绿西瓜。其间十一二岁的少年闰土,项带银圈,手捏钢叉,向一匹猹刺去。那猹却将身一扭,反从他胯下逃走——这幅月夜刺猹的画面,成了“我”三十年来难忘的剪影。那年,因家中轮到三十余年一遇的大祭祀值年,祭器贵重需人看管,父亲便允了忙月的请求,唤其子闰土进城相助。 | |
| "Do not go gentle into that good night" is a poem in the form of a villanelle by Welsh poet Dylan Thomas (1914–1953), and is one of his best-known works. Though first published in the journal Botteghe Oscure in 1951, Thomas wrote the poem in 1947 while visiting Florence with his family. The poem was subsequently included, alongside other works by Thomas, in In Country Sleep, and Other Poems (New Directions, 1952) and Collected Poems, 1934–1952 (Dent, 1952). The poem entered the public domain in all countries outside the United States on 1 January 2024. It has been suggested that the poem was written for Thomas's dying father, although he did not die until just before Christmas in 1952. It has no title other than its first line, "Do not go gentle into that good night", a line that appears as a refrain throughout the poem along with its other refrain, "Rage, rage against the dying of the light". | Do not go gentle into that good night Old age should burn and rave at close of day Rage, rage against the dying of the light Though wise men at their end know dark is right Because their words had forked no lightning They do not go gentle into that good night |
用户也可将 Qwen3-TTS 创建的音色进行持久存储和重复调用,生成生动自然的多轮次多角色长篇章对话。
| 控制类型 | 控制指令 | 合成文本 | 合成样例 |
|---|---|---|---|
| 音色复用 | "旁白": "声音特征沉稳、客观、略带叙事感的女播音腔,普通话标准,语速适中,带有轻微的环境氛围渲染,语调平缓但富有感染力,在关键情节时稍作停顿,增强画面感。情感冷静旁观,偶尔带一丝微妙的反讽" "小林": "25岁男性上班族,声音清亮但时常犹豫,语速时快时慢,紧张时会轻微结巴。情绪波动明显,从低声呢喃到突然激动再到自我怀疑的叹气。肢体语言丰富,经常无意识的小动作" "御姐": "模拟成熟性感的御姐音色,声音略带磁性且沉稳,语速不快不慢,语调充满自信和一丝挑逗,尾音可以稍微拖长并上扬,给人一种游刃有余的掌控感。" | 旁白: 小林今天第三次走神了。酒吧昏黄的灯光晃得他心跳加速,而吧台对面那个红唇微扬的女人,正用指尖轻轻摩挲着酒杯边缘。 御姐: 小弟弟,有兴趣陪姐姐喝一杯吗? 小林: 啊?我、我……我其实不太会喝酒…… 旁白: 他的手指无意识地抠着杯沿,喉结上下滚动,像被什么无形的东西掐住了呼吸。 御姐: 不会喝?那正好——姐姐教你。这杯莫吉托,甜得刚好,就像你刚才偷看我的眼神。 小林: 我、我没偷看!……好吧,看了一眼。就一眼! 旁白: 他猛地坐直,又立刻缩回肩膀,仿佛那句话烫伤了自己的嘴。 御姐: 紧张什么?你连坐姿都在发抖……要不要靠过来一点?这里太吵了。 小林: 靠过去?可、可我们才第一次见面……你都不认识我…… 御姐: 名字不重要,感觉才重要。......而我感觉……你有点可爱。 旁白: 小林的耳朵瞬间红透,连耳后那颗小痣都像在发烫。他想逃,脚却像钉在了高脚凳上。 小林: 可爱?没人这么说过我……他们都说我太闷,连朋友圈都发不出手…… 御姐: 那现在呢?敢不敢发一条——'今晚,和一个危险又迷人的姐姐喝了一杯'? 小林: ……我连配图都不敢选。你笑起来太……太有杀伤力了。 御姐: 那就别发了。有些故事,只适合藏在两个人的记忆里——比如,接下来你打算请我跳支舞吗? 旁白: 他张了张嘴,没发出声音。但这一次,他没有低头,而是轻轻推开了那杯没动过的苏打水,朝她伸出了手。 | |
| "Lucas": "Male, 17 years old, tenor range, gaining confidence - deeper breath support now, though vowels still tighten when nervous" "Mia": "Female, 16 years old, mezzo-soprano range, softening - lowering register to intimate speaking voice, consonants softening" | Lucas! You dropped your... uh... calculus notebook? I mean, I think it's yours? Maybe? Mia wow, my mortal enemy - Mr. Thompson's problem sets. Thanks for rescuing me from that F. Lucas problem! I actually... kinda finished those already? If you want to compare answers or something... Mia this your sneaky way of saying you want to study together, Lucas? Because I saw you staring during lab partners sign-up. Lucas? No! I mean yes but not like... I just think you're... your titration technique is really precise! Mia's the nerdiest compliment I've ever gotten. Tell you what - help me survive pre-calc and I'll teach you how to actually flirt. Lucas, harsh. And here I thought my titration line was smooth. Mia was adorable. Like when you tripped over your shoelaces in the hall yesterday. Or that time you— Lucas okay! I get it, I'm a disaster. So... library after school? I'll bring the graphing calculators? Mia if you promise not to spill coffee on my notes again... though I guess watching you panic-clean was pretty cute. |
使用方法#
通过 Qwen API 使用 Qwen3-TTS-VD-Flash 是非常简单的,我们在下面提供了一个简单的代码片段以供体验:
import requests
import base64
import os
def create_voice_and_play():
# 新加坡和北京地域的API Key不同。获取API Key:https://help.aliyun.com/zh/model-studio/get-api-key
# 若没有配置环境变量,请用百炼API Key将下行替换为:api_key = "sk-xxx"
api_key = os.getenv("DASHSCOPE_API_KEY")
if not api_key:
print("错误: 未找到DASHSCOPE_API_KEY环境变量,请先设置API Key")
return None, None, None
# 准备请求数据
headers = {
"Authorization": f"Bearer {api_key}",
"Content-Type": "application/json"
}
data = {
"model": "qwen-voice-design",
"input": {
"action": "create",
"target_model": "qwen3-tts-vd-realtime-2025-12-16",
"voice_prompt": "沉稳的中年男性播音员,音色低沉浑厚,富有磁性,语速平稳,吐字清晰,适合用于新闻播报或纪录片解说。",
"preview_text": "各位听众朋友,大家好,欢迎收听晚间新闻。",
"preferred_name": "announcer",
"language": "zh"
},
"parameters": {
"sample_rate": 24000,
"response_format": "wav"
}
}
# 以下为北京地域url,若使用新加坡地域的模型,需将url替换为:https://dashscope-intl.aliyuncs.com/api/v1/services/audio/tts/customization
url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"
try:
# 发送请求
response = requests.post(
url,
headers=headers,
json=data,
timeout=60 # 添加超时设置
)
if response.status_code == 200:
result = response.json()
# 获取音色名称
voice_name = result["output"]["voice"]
print(f"音色名称: {voice_name}")
# 获取预览音频数据
base64_audio = result["output"]["preview_audio"]["data"]
# 解码Base64音频数据
audio_bytes = base64.b64decode(base64_audio)
# 保存音频文件到本地
filename = f"{voice_name}_preview.wav"
# 将音频数据写入本地文件
with open(filename, 'wb') as f:
f.write(audio_bytes)
print(f"音频已保存到本地文件: {filename}")
print(f"文件路径: {os.path.abspath(filename)}")
return voice_name, audio_bytes, filename
else:
print(f"请求失败,状态码: {response.status_code}")
print(f"响应内容: {response.text}")
return None, None, None
except requests.exceptions.RequestException as e:
print(f"网络请求发生错误: {e}")
return None, None, None
except KeyError as e:
print(f"响应数据格式错误,缺少必要的字段: {e}")
print(f"响应内容: {response.text if 'response' in locals() else 'No response'}")
return None, None, None
except Exception as e:
print(f"发生未知错误: {e}")
return None, None, None
if __name__ == "__main__":
print("开始创建语音...")
voice_name, audio_data, saved_filename = create_voice_and_play()
if voice_name:
print(f"\n成功创建音色 '{voice_name}'")
print(f"音频文件已保存: '{saved_filename}'")
print(f"文件大小: {os.path.getsize(saved_filename)} 字节")
else:
print("\n音色创建失败")Qwen3-TTS-VC-Flash#
Qwen3-TTS 支持通过自然3s级别音色克隆,并且可以基于克隆的音色生成多语种音频,同时对复杂文本和野生音频都有较高的鲁棒性。
指标#
多语种音色克隆:在MiniMax TTS Multilingual Test Set上,Qwen3-TTS在中、英、法、意大利等语项的内容稳定性优于MiniMax、ElevenLabs 及 GPT-4o-Audio-Preview;其平均词错误率(WER)位居第一。

样例#
| 克隆类型 | 参考音频 | 合成文本 | 合成样例 |
|---|---|---|---|
| 中英克隆 | 昨夜雨疏风骤,浓睡不消残酒。试问卷帘人,却道海棠依旧。知否,知否?应是绿肥红瘦。 | ||
| Overcome with guilt, Martin hung his head and muttered, "I’m so sorry. I never meant to hurt you like this. Can you ever forgive me?" It was obvious what the answer would be. | |||
| 再说,学好文化搞通思想,道理还不是为了劳动?难道我劳动比谁差! | |||
| Innovation blossoms when we cast aside the paralyzing fear of failure and wholeheartedly embrace the gloriously messy, unexpectedly beautiful journey of creation. | |||
| 多语种克隆 | 上周我去日本旅游,看到一个法国人在买东西,那个日本人问:なんか買うものありますか?那个法国人说:Je voudrais acheter un t-shirt à manches courtes.' | ||
| 呐,跟你说个秘密哦!The tea eggs sold by the old lady on the mountaintop are something else, I tell you! Es wird durch Einkochen mit Quellwasser und drei verschiedenen Wildkräutern zubereitet. | |||
| 复杂文本鲁棒性 | Qwen-TTS 是支持音色克隆、生成、控制的语音合成模型,不仅支持多语言multilingual,还支持各种复杂文本,如pin1 yin1,特殊符号等·〛』];能读出各种生僻字词。快来试试吧! | ||
| Is there anyone who can solve the equation: x = [-b ± √(b²-4ac)] / 2a? Nobody can — it's a disaster, very sad! If you know this formula, please email [email protected]. | |||
| 野生音频鲁棒性 | 妲己凭借着自己的妖娆妩媚,在商纣王的宫廷中弄权,她的行为可谓是牝鸡司晨,加速了商朝的灭亡 | ||
| We began our discussion on the four development phases of romantic relationships by reading a quote from 'The General Theory of Love.' |
想听听动物如何开口说话吗?Qwen3-TTS为你展示下跨物种克隆:
| 克隆类型 | 参考音频 | 合成文本 | 合成样例 |
|---|---|---|---|
| 跨物种克隆 | 这点小事都能办砸?说好晚上七点准时开饭,本汪的肚子都咕咕叫了! | ||
| That was one small leap for me, but a giant leap for goatkind! | |||
| 早起的鸟儿有虫吃,早起的虫儿被我吃! | |||
| Oink... not now. My mud nap is at peak fluffiness. Disturb me and I’ll snore louder. |
使用方法#
通过 Qwen API 使用 Qwen3-TTS-VC-Flash 是非常简单的,我们在下面提供了一个简单的代码片段以供体验:
# DashScope SDK 版本需要不低于1.23.9,Python版本需3.10及以上
# coding=utf-8
# Installation instructions for pyaudio:
# APPLE Mac OS X
# brew install portaudio
# pip install pyaudio
# Debian/Ubuntu
# sudo apt-get install python-pyaudio python3-pyaudio
# or
# pip install pyaudio
# CentOS
# sudo yum install -y portaudio portaudio-devel && pip install pyaudio
# Microsoft Windows
# python -m pip install pyaudio
import pyaudio
import os
import requests
import base64
import pathlib
import threading
import time
import dashscope # DashScope Python SDK 版本需要不低于1.23.9
from dashscope.audio.qwen_tts_realtime import QwenTtsRealtime, QwenTtsRealtimeCallback, AudioFormat
# ======= 常量配置 =======
DEFAULT_TARGET_MODEL = "qwen3-tts-vc-realtime-2025-11-27" # 声音复刻、语音合成要使用相同的模型
DEFAULT_PREFERRED_NAME = "guanyu"
DEFAULT_AUDIO_MIME_TYPE = "audio/mpeg"
VOICE_FILE_PATH = "voice.mp3" # 用于声音复刻的本地音频文件的相对路径
TEXT_TO_SYNTHESIZE = [
'对吧~我就特别喜欢这种超市,',
'尤其是过年的时候',
'去逛超市',
'就会觉得',
'超级超级开心!',
'想买好多好多的东西呢!'
]
def create_voice(file_path: str,
target_model: str = DEFAULT_TARGET_MODEL,
preferred_name: str = DEFAULT_PREFERRED_NAME,
audio_mime_type: str = DEFAULT_AUDIO_MIME_TYPE) -> str:
"""
创建音色,并返回 voice 参数
"""
# 新加坡和北京地域的API Key不同。获取API Key:https://help.aliyun.com/zh/model-studio/get-api-key
# 若没有配置环境变量,请用百炼API Key将下行替换为:api_key = "sk-xxx"
api_key = os.getenv("DASHSCOPE_API_KEY")
file_path_obj = pathlib.Path(file_path)
if not file_path_obj.exists():
raise FileNotFoundError(f"音频文件不存在: {file_path}")
base64_str = base64.b64encode(file_path_obj.read_bytes()).decode()
data_uri = f"data:{audio_mime_type};base64,{base64_str}"
# 以下为北京地域url,若使用新加坡地域的模型,需将url替换为:https://dashscope-intl.aliyuncs.com/api/v1/services/audio/tts/customization
url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"
payload = {
"model": "qwen-voice-enrollment", # 不要修改该值
"input": {
"action": "create",
"target_model": target_model,
"preferred_name": preferred_name,
"audio": {"data": data_uri}
}
}
headers = {
"Authorization": f"Bearer {api_key}",
"Content-Type": "application/json"
}
resp = requests.post(url, json=payload, headers=headers)
if resp.status_code != 200:
raise RuntimeError(f"创建 voice 失败: {resp.status_code}, {resp.text}")
try:
return resp.json()["output"]["voice"]
except (KeyError, ValueError) as e:
raise RuntimeError(f"解析 voice 响应失败: {e}")
def init_dashscope_api_key():
"""
初始化 dashscope SDK 的 API key
"""
# 新加坡和北京地域的API Key不同。获取API Key:https://help.aliyun.com/zh/model-studio/get-api-key
# 若没有配置环境变量,请用百炼API Key将下行替换为:dashscope.api_key = "sk-xxx"
dashscope.api_key = os.getenv("DASHSCOPE_API_KEY")
# ======= 回调类 =======
class MyCallback(QwenTtsRealtimeCallback):
"""
自定义 TTS 流式回调
"""
def __init__(self):
self.complete_event = threading.Event()
self._player = pyaudio.PyAudio()
self._stream = self._player.open(
format=pyaudio.paInt16, channels=1, rate=24000, output=True
)
def on_open(self) -> None:
print('[TTS] 连接已建立')
def on_close(self, close_status_code, close_msg) -> None:
self._stream.stop_stream()
self._stream.close()
self._player.terminate()
print(f'[TTS] 连接关闭 code={close_status_code}, msg={close_msg}')
def on_event(self, response: dict) -> None:
try:
event_type = response.get('type', '')
if event_type == 'session.created':
print(f'[TTS] 会话开始: {response["session"]["id"]}')
elif event_type == 'response.audio.delta':
audio_data = base64.b64decode(response['delta'])
self._stream.write(audio_data)
elif event_type == 'response.done':
print(f'[TTS] 响应完成, Response ID: {qwen_tts_realtime.get_last_response_id()}')
elif event_type == 'session.finished':
print('[TTS] 会话结束')
self.complete_event.set()
except Exception as e:
print(f'[Error] 处理回调事件异常: {e}')
def wait_for_finished(self):
self.complete_event.wait()
# ======= 主执行逻辑 =======
if __name__ == '__main__':
init_dashscope_api_key()
print('[系统] 初始化 Qwen TTS Realtime ...')
callback = MyCallback()
qwen_tts_realtime = QwenTtsRealtime(
model=DEFAULT_TARGET_MODEL,
callback=callback,
# 以下为北京地域url,若使用新加坡地域的模型,需将url替换为:wss://dashscope-intl.aliyuncs.com/api-ws/v1/realtime
url='wss://dashscope.aliyuncs.com/api-ws/v1/realtime'
)
qwen_tts_realtime.connect()
qwen_tts_realtime.update_session(
voice=create_voice(VOICE_FILE_PATH), # 将voice参数替换为复刻生成的专属音色
response_format=AudioFormat.PCM_24000HZ_MONO_16BIT,
mode='server_commit'
)
for text_chunk in TEXT_TO_SYNTHESIZE:
print(f'[发送文本]: {text_chunk}')
qwen_tts_realtime.append_text(text_chunk)
time.sleep(0.1)
qwen_tts_realtime.finish()
callback.wait_for_finished()
print(f'[Metric] session_id={qwen_tts_realtime.get_session_id()}, '
f'first_audio_delay={qwen_tts_realtime.get_first_audio_delay()}s')引用#
如果你觉得我们模型在您的研究中很有效果,请考虑引用我们 📝:)
@misc{qwen3_tts_202512,
author = {Qwen Team, Alibaba},
title = {Qwen3-TTS Steps Up: Voice Cloning and Voice Design!},
year = {2025},
url = {https://qwen.ai/blog?id=qwen3-tts-vc-voicedesign},
urldate = {2025-12-23}
}