Yes, you read that right. In what stands as a testament to the gains made by speech recognition technology, YouTube has announced that it has automatically captioned over 1 Billion videos on its platform. The company also disclosed that videos with captions are now watched over 15 million times, everyday.
In case you are unaware of it, the automatic captioning is powered by Google’s speech recognition technology — which is one of the best in the land. Automatic captioning actually reached YouTube way back in 2009. However, the technology was still in its infancy and the captions had a lot of error. Now though, thanks to Google’s machine learning smarts, the error rates have dropped drastically and Google says, that they are nearly on the verge of reaching human capabilities.
The company also said that it had achieved a 50 percent increase in accuracy as a result of the efforts made over the past few years.
Meanwhile, YouTube actually is in a very good position to train a voice recognition and captioning service. The platform is home to billions of videos with more added every single day. As such, a huge assortment of training data is available to the company along with the community, which acts as moderator and reviews the transcript. What’s more, content is available in multiple languages, allowing for a significantly broader dataset than that available elsewhere.
The advantage of this community powered review process is realized when the human corrected captions flow back into the main data base. Since its all machine learning, the system learns from the errors pointed out by the community and does not repeat them in the future.
So while captioning its content, YouTube is also helping Google’s automatic captioning service become better. And since it is Google, you can bet that the improvements made over here will also make their way to the other products belonging to the company.
