AI transcription so training videos hit home

Cloudinary’s challenge

Neither human nor AI transcription services were able to accurately transcribe Cloudinary’s training videos, which were loaded with technical terms, proprietary lingo, and acronyms (and frequently taught by non-native English speakers). Correcting the transcripts took hours, and blocked the Cloudinary team from doing what they wanted to be doing: making more, better videos to help their customers.

AI transcription you can train

Descript automatically transcribes video in seconds, with industry-leading accuracy. The Cloudinary team uses Descript’s transcription glossary feature to teach the AI those especially tricky terms and acronyms, so when it hears “JSON file,” Descript transcribes it that way, not as “Jason file.” Descript’s AI transcription also learns and gets better the more you use it. So after a few months, the Cloudinary team noticed they no longer had to correct one of their most common, important, and mistranscribed words: “Cloudinary.”

Of course, no transcription — human or AI — is or ever will be perfect; Descript makes the inevitable manual corrections easier too. The Cloudinary team can invite multiple editors to every video project, so instructors and subject-matter experts can quickly scrub inaccuracies using Descript’s effortless transcript-correction tool. All the changes happen in the cloud, in real time, so editors can work simultaneously, making corrections and leaving comments as they go.

The Cloudinary team came to Descript for better transcription. Then they discovered magical features they never expected, or ever dared to wish for — and transformed their workflow.

Filler word removal – “ums” and “uhs,” gone in a few clicks

Many of Cloudinary’s instructional videos are recorded as livestreams. Because they’re humans, the instructors use a lot of “ums,” “uhs,” “likes,” “you knows,” and other filler words. Most of those were okay for the video — the team wanted to keep it similar to the live version. But removing them from the transcript during editing was a tedious, time-consuming process for the production team.

Until Descript came along. Descript’s automatic filler-word detection and removal enabled Cloudinary’s editors to remove every unwanted filler word from the transcript in a few clicks. Same for any excessive filler words in the video. Descript also adds room tone automatically to smooth the cuts, and the option to restore them quickly where removal feels unnatural.

Regenerative AI for studio-quality audio wherever the experts are

You can send quality microphones to your subject matter experts, but you can’t be sure they’ll set them up right, or that the room they record in won’t be full of reverb, or that their neighbor’s leaf blower won’t start up midway through recording. But clean, clear audio is critical to helping customers comprehend the instruction they’re getting. So fixing those kind of audio challenges used to be a big problem for Cloudinary’s team.

Descript’s Studio Sound solved it almost instantly. It uses AI voice re-generation to strip out background noise, reverb and other stuff you don’t want, then re-construct the voice audio so it sounds like it was recorded in a studio. Studio Sound works so well that Sam Brace no longer uses an external mic to record the Cloudinary podcast — he gets better sound by using his laptop mic, then applying Studio Sound in post-production.

AI voice re-generation — for audio that’s divine