谷歌推YouTube视频字幕自动生成功能

谷歌推YouTube视频字幕自动生成功能

From now on, when you put a video up on YouTube it will get captions.

"This is huge. It's what I dreamt of for years," said Ken Harrenstein, a software engineer at Google who is deaf. Harrenstein announced the new auto-caption software at YouTube's headquarters in San Bruno.

Harrenstein, a graduate of MIT, explained how he often missed lectures when he was in college because he couldn't understand them. Now, many lectures are video taped and put on YouTube. Harrenstein showed how he could view a lecture with captions. But, even more exciting he can search a phrase like "elephant and mouse," and find the exact moment in the lecture where the professor said those words.


Auto-captioning technology isn't new but YouTube engineers say its never been deployed on such a wide scale. Every minute, 20 hours of video are uploaded to YouTube. Mike Cohen, a research scientist at Google, says the technology has been around for 50 years. But the progress on it has been incremental. Now it's hit a point where it's gotten really useful.

It's useful, but Cohen and Harrenstein admit it's still flawed. They showed a video where the word "salmon" had been substituted for "sim card." They also said that it will be used only on videos with clear enough audio -- that means those folks talking with a stereo in the background and rambling on with loud traffic noise behind them won't get transcribed.

Still, this is a big step. There were a group of students at the announcement from a local deaf school who were absolutely elated about the technology. They were a hip looking group of young women dressed in stylish jeans and shoes and no doubt they were going to feel hipper now that they could watch more YouTube videos without having to keep asking a person who can hear to explain it.

Harrenstein says they will be constantly improving the technology and will hopefully have it available soon to do transcriptions in languages other than English

谷歌去年11月便面向部分视频推出了字幕自动生成功能。该功能利用语音识别技术自动生成字幕，帮助聋哑人了解视频中的对话。目前，谷歌将把该功能推广到所有的YouTube英语视频中。

谷歌软件工程师肯·哈伦斯坦(Ken Harrenstein)表示，配合着谷歌的翻译功能，还可以将英语字幕翻译成其他语言，让全球各地的聋哑人都能够观看YouTube视频。但他也指出，自动字幕生成功能目前还不完善，例如它会把“SIM card”(SIM卡)听成“salmon”(三文鱼)。

从周四开始，所有使用英语的YouTube视频都可以享受到这一功能。但由于技术存在限制，因此需要提供清晰的音轨，如果含有背景噪音或者发音不清，便无法识别。哈伦斯坦表示，美国总统奥巴马最近对智利地震发表的演讲就是一个非常清晰的例子，可以自动生成字幕。

由于技术不完善，因此视频所有者需要对字幕进行核对，以确保精准。但谷歌承诺会继续改善这一技术。以前上传的视频，只要符合上述要求，同样可以使用字幕自动生成功能。

This is a revolution in IT industry.