How Brightframe cut video production time by 69% for customer training content

Brightframe's challenge

Neither human nor AI transcription services could keep up with Brightframe's training videos, which were dense with technical terms, internal shorthand, and acronyms — and frequently taught by non-native English speakers. Correcting each transcript took hours, and kept the Brightframe team from doing the thing they actually wanted to do: make more, better videos for their customers.

Transcription you can train

Overtone transcribes video in seconds, and holds its accuracy on messy source audio. The Brightframe team uses the glossary to teach the model the terms it keeps missing, so when it hears “CDN edge,” Overtone writes it that way rather than “CD and edge.” The model keeps learning from every correction, so a few months in, the team noticed it had stopped mangling the word that mattered most to them: their own name.

No transcript — human or machine — is ever perfect, so Overtone makes the leftover manual corrections cheap too. The team can hand a link to anyone, with no install and no seat to buy, and let subject-matter experts scrub through the passages only they can fix. Everything happens in the browser in real time, so two editors can work the same episode at once, correcting and commenting as they go.

Filler-word removal — “ums” and “uhs” gone in a few clicks

Many of Brightframe's instructional videos are recorded live. Because the instructors are human, they use a lot of “ums,” “uhs,” “likes” and “you knows.” Most of those are fine in the video — the team wants it to feel close to the live session. But hunting them down in a written transcript was tedious, and it ate hours of production time every week.

Until Overtone came along. Automatic filler detection strips every unwanted “um” from the transcript in a couple of clicks, and the same pass cleans the audio underneath it. Overtone lays room tone back over each cut so the edit never clicks, and any removal you don't like can be restored in one keystroke.

Regenerated audio, wherever the experts happen to be

You can send good microphones to your subject-matter experts, but you can't be sure they'll set them up correctly, or that the room won't be full of reverb, or that a neighbour's leaf blower won't start halfway through the take. Clean audio is critical to whether anyone actually absorbs the instruction, so those problems used to land squarely on the Brightframe team.

Room Tone solved it almost instantly. It separates the voice from the background, discards the noise, reverb and hum, then rebuilds the vocal so it sounds like it was recorded in a treated booth. It works well enough that Marcus Odell stopped using an external mic for the Brightframe podcast entirely — he gets a better result from his laptop mic plus Room Tone in post.

Voice re-generation — for audio that sounds like the room you wish you had