Facebook
X
Instagram
Youtube
Sign in
News
Tech today
Android
iPhone
Windows Phone
Apple
Geeks
Linux
Applications
More
Windows Phone
Android
iPhone
Accessories
Bluetooth
Sign in
Welcome!
Log into your account
your username
your password
Forgot your password?
Password recovery
Recover your password
your email
Search
Sign in / Join
Events
Guides
Advertise
Buy Now
Facebook
X
Instagram
Youtube
Sign in
Welcome! Log into your account
your username
your password
Forgot your password? Get help
Password recovery
Recover your password
your email
A password will be e-mailed to you.
News
Tech today
iPhone
Kim Kardashian Shows Off Deep Cleavage In Plunging Top & Mini
iPhone
Thinklab – Building a startup team to fix science and government
iPhone
Sandbox to try out the code written in almost all languages
iPhone
OneNote for iOS and Mac lets you attach files
iPhone
We Found the Sexiest Lingerie on the Internet
All
Android
iPhone
Windows Phone
Apple
Apple
Cheryl Steals Kate Middleton’s Beauty Icon Status
Apple
Beginner: Are you stuck in programming should not do
Apple
50 Tips and Insights About Productivity, Happiness, and Life
Apple
Wind and solar power are even more expensive than is commonly thought
Apple
Android L Will Keep Your Secrets Safer
Geeks
Geeks
50 Tips and Insights About Productivity, Happiness, and Life
Geeks
StreetScore scores a street view based on how safe it looks to a human
Geeks
Going Beyond Amazon: A New Model for Authors, Retailers, and Publishers
Geeks
UK to allow driverless cars on public roads in January
Geeks
How to drive growth through customer support
Linux
Linux
Why you should choose Microsoft over Linux
Linux
Building an API in 60 seconds, without any server setup
Linux
Building a Gimbal in Rust: An Introduction
Linux
Show HN: Appsites – Beautiful websites for mobile
Linux
Xbox One to launch in China this month after all
Applications
Applications
Moogle Corp: Company you might be working for
Applications
The 10 Runway Trends You’ll Be Wearing This Year
Applications
A first glimpse at Java 9: Early access release of JDK9 on OpenJDK
Applications
How Internet Providers Get Around War Zones
Applications
Marriott Plays With Sensory-Rich Virtual Reality Getaways
More
Windows Phone
Android
iPhone
Accessories
Bluetooth
FYIPC
Accessories
AI工具
Apple
Applications
Bluetooth
Cooling
CPU/Motherboard
Geeks
Graphics Card/Monitors
Keyboards/Mice
Laptop
Linux
Memory
Networking
Storage
Tech today
在线工具
学习资源
开发工具
未分类
软件应用
Latest
Latest
Featured posts
Most popular
7 days popular
By review score
Random
ADATA XPG Lancer RGB 64GB DDR5-6000 CL30 Review: The Right Speed, the Wrong Price
frank5
-
2026年5月2日
0
ADATA XPG LANCER RGB 64GB DDR5-6400 CL32 — Performance, Overclocking, and Value Review
frank5
-
2026年1月28日
0
NVIDIA Reportedly Abandons SOCAMM, Pivots to New SOCAMM2 Memory Standard
frank5
-
2025年9月15日
0
Memory
Home
Memory
Memory
G.SKILL Unveils DDR5-8000 CL36 Memory with AMD EXPO Support
frank5
-
2025年3月12日
0
Memory
SK Hynix Achieves 70% Yield Rate for 12-Layer HBM4 Memory in...
frank5
-
2025年2月28日
0
Memory
Micron Introduces 1γnm DDR5 Memory: First EUV Implementation Enables 128GB/9200MHz Performance
frank5
-
2025年2月27日
0
Memory
SK Hynix and Micron to Potentially End DDR4/DDR3 Production, Shifting Focus...
frank5
-
2025年2月20日
0
- Advertisement -
APPLICATIONS
Advanced Shader Delivery Tested: Is PC Gaming Finally Killing Shader Compilation Stutter?
frank5
-
2026年5月25日
An Expert-Level Analysis of the Gigabyte Aorus GeForce RTX 5070 Ti Master Graphics Card
frank5
-
2025年9月17日
Moogle Corp: Company you might be working for
frank5
-
2026年8月14日
How to drive growth through customer support
frank5
-
2026年8月14日
HOT NEWS
微软开源 VibeVoice:一次生成90分钟多人语音 声音越来越像真人了
frank5
-
2026年7月28日
0
AI工具
如果你经常听播客、有声书,或者做视频配音,应该会发现一个问题:AI 配音已经能读出来,但离真正像人在聊天,还有一段距离。 很多语音模型短句表现不错,可一旦让它连续说几分钟,就容易出现声音僵硬、语气重复、角色混乱的问题。尤其是多人对话场景,AI 往往很难保持每个人的声音特点。 微软开源的 Microsoft VibeVoice,就是冲着这个问题来的。它不是简单的文字转语音工具,而是一套面向长文本、多角色对话的语音 AI 模型。项目定位是“开放式语音 AI”,包含文本转语音(TTS)和语音识别(ASR)两大方向。GitHub项目地址:https://github.com/microsoft/VibeVoice最吸引人的地方,是它能生成非常长的连续语音。 传统 TTS 模型通常更适合生成几句话,比如导航播报、短视频配音。但 VibeVoice-TTS 可以根据一篇长文章生成接近完整节目的音频,支持最长约 90 分钟的语音生成,并且可以让最多 4 个不同角色参与对话。简单理解,就是你给它一份访谈稿,它可以直接变成一档多人聊天节目。 这个能力对于内容创作者来说挺有想象空间。 比如你有一篇公众号文章,不想自己录音,可以让 AI 转成播客;一本电子书,可以快速生成有声版本;甚至一份产品介绍,也能变成两个 AI 主持人的讨论节目。 当然,VibeVoice 的重点并不是“把文字念出来”,而是让声音更接近真实交流。 它采用了一套针对长语音设计的方法,把声音拆成更容易处理的语音信息,再结合类似大语言模型理解上下文的方式生成语音。这样做的好处是,AI 不只是关注当前一句话,而是能记住前面的内容,让整个长对话保持连贯。除了 TTS,VibeVoice 还有语音识别模型...