Partnership launches One Click to Caption solution; Provides high quality, fast, and low cost method to caption online media.
cielo24 uses a combination of machines and a crowd-sourced work force to create captions that are 99 percent accurate for $1.50 a minute with a 48-hour turnaround time, with even quicker times and cheaper prices available when accuracy isn’t as critical. Though only seven months old, cielo24 is already serving hundreds of customers – including online education giants Coursera and Cartpinteria-based lynda.com – and handles about 50,000 minutes of video per month.
cielo24 eventually hopes to caption all kinds of professionally created online video, whether it’s breaking news, corporate training materials or educational courses. But the online education space has been an early adopter: That industry saw $1 billion in investment last year and is growing fast, but now faces federal accessibility mandates that require captions. “Just like there have to be ramps in buildings, the government now says you have to have equal access to commercially available media,” said Brian Plackis Cheng, the company’s CEO and a veteran of MCI, Software.com and several South Coast startups.
cielo24’s system can integrate directly with a platform like Coursera, so that when professors or schools upload content, captioning can be selected with a click. “We grab that, caption it, and turn it around in 24 to 48 hours,” Plackis Cheng said.
Cheap computing and storage power and crowdsourcing make it all possible. When a video is uploaded, it’s sent to web servers where speech recognition software detects normal breaks in the speakers voice to split the video up into 30-second chunks. Those chunks are then sent out to platforms such as Amazon’s Mechanical Turk. There, a broad base of users can pick up piecework transcribing the 30-second chunks. (cielo24 taps a range of crowdsourcing platforms to produce captions in 10 languages.) Crowd workers have to prove their accuracy and get paid more as they get better, eventually overseeing other new users.
That gets the transcriptions to 95 percent or so accurate. From there, the content gets sent to facilities in Africa or the Philippines that function as a call-center-like sort of captive crowd. Those facilities have the skills to bring the transcripts up to 99 percent accurate, and also serve as a fallback in case the more openly sourced crowd workers can’t meet turnaround requirements. The idea is that you don’t need deep expertise until the final 5 percent of the end product. Once all the work is done, computers stitch the final product back together. The result is something much more powerful than a plain transcript because there’s also timestamps and other data associated with each word. “You can search down to the frame in the video where the word is said,” Plackis Cheng said. “It’s the only way we’re going to be able to deal with the deluge of video on the Internet.”
Crowdsourcing makes it possible to provide lower prices and quicker turnaround than traditional services, but slicing and dicing the videos also makes shorter videos economical. The vast majority of lessons in online education are being broken up into segments of less than 10 minutes. “Most transcription companies don’t want the brain damage of dealing with little videos. Our system is indifferent,” Plackis Cheng said.
Andrew Tompkins, senior closed caption editor at lynda.com, said the change has been huge for his company. “We want to get all of our movies captioned for accessibility but also as a search tool for new members. The main thing for us is that cielo24 can turn around content really fast,” Tompkins told the Business Times. “Before, we were at about two to three weeks, and now we’re at about two to three days:’
While lynda.com has broadened its course offerings over the years, it still retains a lead in showing digital professionals all the quirks of the latest software packages. Fiona Trayler, director of production and post production at the firm, said one of the most frequent feedback questions from users was when a new course would be ready. “It’s really helped us with what we were trying to achieve with our time to market,” Trayler said.
For now, cielo24 is getting a boost from accessibility requirements, which may also become apart of all public proceedings in California. The firm already handles captioning for Santa Barbara City College’s trustee meetings. But the longer-term goal is addressing the billions of minutes of high-quality professional content that’s technically available but not particularly useful because users have to wade through it to find anything. “That video is not searchable. It’s a black box online,” Plackis Cheng said.
The possibilities are broad. One could be helping holders of old-line content like classic TV shows create chances to broaden their audience by helping a fan stumble onto a searched catchphrase. Another would be providing a better foundation for contextual video advertisements. Right now, those ads are based on the text you searched to find a video, not what you’re actually watching. “Our goal isn’t to be the search engine. We want to create the data,” Plackis Cheng said.