When Google rolled out its automatic captioning service for YouTube back in 2009, the results were far from great. Since then, Google’s speech recognition technology has made massive strides, though, and while the service still makes its fair share of mistakes, it’s now quite good at making videos accessible to the deaf and hearing impaired. As the YouTube team announced today, the service has now captioned one billion videos and people watch video with captions 15 million times a day.
Google says the service is getting closer and closer to human transcription error rates and that all the various improvements the company has made over the years have led to a 50 percent increase in accuracy. The wide variety of content on YouTube obviously makes it hard to caption everything, but it also offers the company with a wide range of training data. When the YouTube community reviews and edits these automatic captions, that information then flows back into the machine learning process to make the system more accurate.
The service’s automatic captions currently support 10 languages. YouTube program manager Liat Kaver today writes that the company plans to roll out many of the improvements it has recently made to these additional languages, too.