Side-by-side voice demos, built for focused listening

Audio Comparison Lab

Compare the reference prompt, CosyVoice 2.0, CosyVoice 3.0-0.5B, and CosyVoice 3.0-1.5B in one focused listening table. Every player streams the original WAV demo directly in the browser.

Showing 31 rows. Audio uses preload="none" so the page stays light until a listener presses play.

11 comparisons

Zero-shot In-context Generation

Multilingual zero-shot cloning across normal and hard pronunciation samples.

ZHreference prompt

转任福建路转运判官。

synthesis text

这个记者俱乐部将成为该部门发放资讯的唯一渠道。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

HARD-ZHreference prompt

在中国鸦片泛滥的年代,不同材质的烟枪甚至成为了身份和地位的象征。

synthesis text

八百标兵奔北坡,北坡炮兵并排跑,炮兵怕把标兵碰,标兵怕碰炮兵炮。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

ENreference prompt

There is no lock but a golden key will open it.

synthesis text

Butler tranquilizes Holly with a hypodermic dart gun.

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

HARD-ENreference prompt

And there were dunes, rocks, and plants that insisted on living where survival seemed impossible.

synthesis text

Fuzzy Wuzzy was a bear. Fuzzy Wuzzy had no hair. Fuzzy Wuzzy wasn't very fuzzy, was he?

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

JAreference prompt

来週、美容院で髪を切ろうと思っています。

synthesis text

結合することができないから矛盾するというのである。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

KOreference prompt

그들이 집까지 왔을 때는 어슬어슬한 황혼이었다.

synthesis text

동혁은 환자의 머리맡을 떠나지 않았다.

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

DEreference prompt

Zieht euch bitte draußen die Schuhe aus.

synthesis text

Die Gemeinde liegt im Westerwald im Einzugsbereich der Westerwälder Seenplatte.

CosyVoice 2.0

Previous generation baseline.

Not supported

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

ESreference prompt

Durante unos años, enseñó Física e Historia en el colegio de nobles de Parma.

synthesis text

En el horizonte, un sol radiante, símbolo del nacimiento de la nación.

CosyVoice 2.0

Previous generation baseline.

Not supported

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

FRreference prompt

Ce dernier a évolué tout au long de l'histoire romaine.

synthesis text

De manière informelle, la capacité d'un modèle de classification correspond à sa complexité.

CosyVoice 2.0

Previous generation baseline.

Not supported

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

ITreference prompt

Fin dall'inizio la sede episcopale è stata immediatamente soggetta alla Santa Sede.

synthesis text

Il suo primo utilizzo è stato in Vietnam da parte dell'esercito statunitense.

CosyVoice 2.0

Previous generation baseline.

Not supported

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

RUreference prompt

Неожиданно катастрофа приобрела глобальные масштабы.

synthesis text

Таким образом, мы услышим от них более сфокусированные и полезные выступления.

CosyVoice 2.0

Previous generation baseline.

Not supported

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

5 comparisons

Mixed-lingual In-context Generation

One reference voice speaking Chinese, English, Japanese, and Korean mixed in the same sentence.

Sample 1reference prompt

今天我们看到模型的本质,其实在很多时候是今天把我们人类的知识能够有效的聚集起来。能够成为今天我们一个重要的一个智慧体,去帮助我们的业务的开发,帮助我们去解决行业的知识。今天我们实际上是从一个信息时代,真正进入到了一个智能时代。在智能时代里面一个重要的一个环节就是模型变得无处不在,也是模型所代表的知识体系。

synthesis text

听客户反馈说我们的cosyvoice的声音复刻能力,音色相似度比较高,我也来试一试,大家听听这句话像我说的吗?

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Sample 2reference prompt

今天我们看到模型的本质,其实在很多时候是今天把我们人类的知识能够有效的聚集起来。能够成为今天我们一个重要的一个智慧体,去帮助我们的业务的开发,帮助我们去解决行业的知识。今天我们实际上是从一个信息时代,真正进入到了一个智能时代。在智能时代里面一个重要的一个环节就是模型变得无处不在,也是模型所代表的知识体系。

synthesis text

晚上去超市买了很多水果,还有一些美味的アイスクリーム。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Sample 3reference prompt

今天我们看到模型的本质,其实在很多时候是今天把我们人类的知识能够有效的聚集起来。能够成为今天我们一个重要的一个智慧体,去帮助我们的业务的开发,帮助我们去解决行业的知识。今天我们实际上是从一个信息时代,真正进入到了一个智能时代。在智能时代里面一个重要的一个环节就是模型变得无处不在,也是模型所代表的知识体系。

synthesis text

今天晚饭吃了韩国泡菜찌개,味道真的很好。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Sample 4reference prompt

今天我们看到模型的本质,其实在很多时候是今天把我们人类的知识能够有效的聚集起来。能够成为今天我们一个重要的一个智慧体,去帮助我们的业务的开发,帮助我们去解决行业的知识。今天我们实际上是从一个信息时代,真正进入到了一个智能时代。在智能时代里面一个重要的一个环节就是模型变得无处不在,也是模型所代表的知识体系。

synthesis text

I can't wait to try the new 비빔밥 restaurant downtown this weekend!

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Sample 5reference prompt

今天我们看到模型的本质,其实在很多时候是今天把我们人类的知识能够有效的聚集起来。能够成为今天我们一个重要的一个智慧体,去帮助我们的业务的开发,帮助我们去解决行业的知识。今天我们实际上是从一个信息时代,真正进入到了一个智能时代。在智能时代里面一个重要的一个环节就是模型变得无处不在,也是模型所代表的知识体系。

synthesis text

Tomorrow, I'm planning to go shopping for a new テレビ at the electronics store.

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

10 comparisons

Emotionally Expressive Voice Generation

Happy, sad, fearful, angry, and surprised delivery in English and Chinese.

Happyreference prompt

Great, yeah. I mean, it has been great, too. You know, some of these people must have seen me play before because they were requesting a bunch of my songs.

synthesis text

I actually managed to grab tickets to Eason Chan’s concert! It’s amazing! I can’t wait to hear his voice live!

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Happyreference prompt

终于去看运动会啦,舒畅啊!

synthesis text

等到七月底项目结束,我就可以申请休年假了,好期待哦!

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Sadreference prompt

Born once every 100 years, dies in flames.

synthesis text

The research institute won’t renew my contract, so I have to go back to my country. I don’t even have a job anymore.

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Sadreference prompt

红了鼻头的小丑,眼泪止不住的流,流到嘴边咽下悲伤。

synthesis text

原来父皇什么都知道了,我真笨。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Fearfulreference prompt

I... I'm really nervous about getting my hair cut here... What if it doesn't turn out the way I want? I... I don't know if I can go through with it.

synthesis text

I’m too scared to be here alone, can you stay with me?

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Fearfulreference prompt

不断进步的科技,是不是会让医生不再需要人类来担任呢?

synthesis text

求你了,别再说那些吓人的话,我真的快撑不住了。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Angryreference prompt

The boy, O'brien, was specially maltreated.

synthesis text

I told you, stop nagging, alright? Haven’t you nagged enough in this lifetime?

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Angryreference prompt

受到处罚你可不能怨别人,知道吗,臭小子!

synthesis text

老喜欢出阴招,使手段,太不厚道了!

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Surprisedreference prompt

I can't believe it— the lions just broke out of their enclosure and are walking around freely!

synthesis text

What did you just say? Is this really true? I can’t believe it at all!

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Surprisedreference prompt

真的吗?!每个人居然真的都有权利追求自己的幸福?!这真是太不可思议了!

synthesis text

我怎么听着像在做梦一样,怎么可能会发生这样的事?

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

5 comparisons

Chinese dialect Voice Generation

Dialect transfer examples including Cantonese, Dongbei, Tianjin, Sichuan, and Shanghainese.

Cantonesereference prompt

但系,好明显唔系啦。

synthesis text

各方面令到一啲行业咧嗰个营运咧系受较为大影响嘅,比如教育啊,即系补习社啊。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Dongbeireference prompt

我媳妇说:“啥?玩愣?你说啥?我没听清,你再说一遍。”

synthesis text

但是,没事的时候呢,我们一起来回忆回忆,想一想过去的我们。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Tianjinreference prompt

就问问,这锣是哪儿的人告诉是天津的。

synthesis text

地方志都没有,就凭脑子记,有些不是那么准确的,有些个就遗忘。这就随时代呀,都都在改变。嗯,这是很正常的事。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Sichuanreference prompt

此次新增的两列车,是整个增车项目的首批。

synthesis text

目前,由于小张不配合,导致交警没有对事故划分责任。

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.

Shanghaireference prompt

没钞票侬凭啥爱我?

synthesis text

侬为啥背上炸药包?

CosyVoice 2.0

Previous generation baseline.

CosyVoice 3.0-0.5B

CosyVoice 3 base output.

CosyVoice 3.0-1.5B

Large model output for quality comparison.