Show HN: Edit videos faster by automatically removing silences (kapwing.com)
Our team is filled with technologists and creators, and when we record and edit videos, 80% of the time is spent chopping up the video, removing silences, and picking the right takes. So we decided to build a tool that did that for you — or at least get you there most of the way!
Our initial implementation is somewhat naïve and uses a user configurable silence threshold that just reads in volume levels. In the future, we’d like to use a frequency-based approach that focuses on the human voice. We’re also open to ideas, so let us know if you have any!
93 comments
[ 3.6 ms ] story [ 162 ms ] threadI've been watching a lot of educational videos and lectures on 2x or even 3x speed, thereby saving a ton of time.
Of course, when the entire video is watched at 2x speed, the pauses can comfortably be shown at 5x speed or even faster.
The author also wrote supporting scripts for Vegas to extract scenes based on the position of the blips in the audio.
Anecdata - heavily jump-cut edits on youtube instantly inspire me to find an alternative source of the same information, as the breathless, rapid-fire, jerky-video, sensation is irksome. Evidently I'm in the minority, and I'm okay with that.
Not exactly. Most movies use https://en.wikipedia.org/wiki/Single-camera_setup, so even though there might be cuts to different camera angles, each angle actually comes from different takes. The goal of different angles isn't really to cut out pauses/fillers, it's for stylistic/cinematography reasons.
>People started doing jump cuts in youtube videos and for some reason we're stuck with it.
Probably because setting up multiple cameras and lighting them is cost prohibitive for low budget videos on youtube?
Yes. One classic example is in a scene with two people speaking, you have two different camera angles to switch focus between them.
More generally, film makers who want to focus / depth-of-field effects very often do so with two cameras / takes where the focus is already set. Changing focus in the middle of a shot is very jarring.
Incidentally, this is something that became immediately obvious to me when I started to use my DSLR for video.
At the top end you've got people comfortable with one or more cameras, a clear message, good communication skills, a well honed script / presentation -- in those cases the information density is naturally going to be high, and any brief pauses are thoughtfully placed to allow the viewer to contemplate & digest. These types of videos are rare but excellent, per my taste.
As I noted, I'm definitely in the minority camp here --the desire that a poorly prepared video should have been a web page is probably generationally akin to the comparably curmudgeonly contemplation that rambling online meetings should have been an email.
A more complex algorithm might watch movement and tune the amount of speed up so the visual doesn't get jumpy.
It’s Mac-only right now but I’m currently working on a Windows version.
I’m a big fan of one-time licenses and the Sketch-style “1 year of updates but use it forever if you want” kind of thing too, and Recut will probably switch to that at some point to make this whole thing more sustainable. What’s interesting is I’ve noticed that not everyone feels this way - some people genuinely would rather pay a smaller fee, but monthly. The thought has crossed my mind to offer both.
I do think there’s a large chunk of the market who don’t use NLEs and would rather avoid them (especially in the mobile crowd) and something web-based like Kapwing is probably preferable for those folks.
My first impression when I checked out the landing page and product was: this is a perfect bootstrapped SaaS product idea to grow to meaningful ARR with a small team of 2-3 + some hired help for things like customer support.
tl;dr is that 1) It's a competitive domain, so we felt we needed to grow our team quickly to stay relevant 2) There's a lot of technical challenges with video editing in the browser + cloud, so we needed to be able to hire talented engineers to scale and 3) I believe there's a huge business in empowering creative teams, so we think we can get to venture scale and needed the capital to support that.
You can read what they're saying in a fraction of the time it takes to watch a video, and often internalize it better.
Doing it natively was hard enough, and recently I've been rewriting Recut with Rust + Electron so I have an idea of how much work it was to get it working well :sweat-smile: Keep up the good work y'all!
On the Rust/Electron side, all the heavy lifting is done in Rust-land, and so far, it's going pretty well. It's not all roses of course - it takes a bounce or two to start instead of launching instantly, and it seems there's not much to be done about Electron's baseline memory usage or size on disk.
(I went down a crazy rabbit hole of compiling Chromium/Electron to see if there were things that could be stripped out, because I honestly don't need much, but realized that is a very large undertaking. Possibly another day. Some kind of "Electron Lite" would sure be awesome...)
But in general I'm spending an inordinate amount of time paying attention to performance. It was my main concern with the Electron platform. So far I've got playback on par, and stuff like loading files, silence detection, and drawing waveforms is actually faster than the Mac version. It feels snappy.
I looked at a bunch of cross-platform UI toolkits, and everything has its tradeoffs. Qt is probably the most full-featured alternative, and I’ve had experience with it in the past, but I didn’t love the idea of tackling a project this size in C++. On the Rust side, Tauri is very promising but it’s still early days.
Aside from the framework, the other big tradeoff is 1 app vs multiple - more “nativeness” for a much slower feature velocity. The big companies like Slack and Figma don’t even make that trade with millions in funding… and it seemed unwise to take that on as a solo bootstrapped developer.
In terms of stuff like ecosystem, building/packaging, updater support, and even nice little native details like progress bars on app icons, Electron has a lot going for it. The big downsides in my mind are startup time, baseline memory usage, and disk size. Once it’s up and running, IMO app performance is almost all down to the app’s code. So I’m optimizing where I can - being mindful of algorithms, avoiding memory copies, trying to keep things cache-friendly, avoiding heavy JS stuff, taking copious benchmarks, etc.
[1] https://tauri.studio/
It would help with both the disk and memory usage, easy to learn after web frameworks (both the framework, Flutter, feels familiar and Dart is super easy to learn), works well with Rust, but of course it has its downsides (no longer can rely on the html js css ecosystem).
IMO this project is already in debt on innovation tokens[0] given I didn't know Rust coming in, nor much at all about video. And then, hybrid web/native stuff is just not very widespread so there's not a lot of existing answers for things. Lots of digging into code and figuring stuff out. The web stack (Svelte, TypeScript) is the one part of this thing that already felt familiar, and I didn't want to throw that out.
0: from Choose Boring Technology :) http://boringtechnology.club/
However, recently I updated my recording setup with a new platform. Now I have separate video tracks for each participant and recut doesn’t handle multiple videos. Will this be supported in the new release?
I’ve heard it said before (I think it was on the Software Social podcast) that whenever you have a choice to support 1 or many of a thing, always err on the side of “many” (users, teams, etc.) Well it turns out that applies to video tracks and clips too! There are massively more things to consider once there’s more than one, haha.
On 1 vs many,I think its totally reasonable to start with 1 given you are a small team. If you force yourself with high standard (which takes lots of extra person-hours), then you lose the key agility compared with other bigger players.
This one in particular is the one where I was introduced to the style but it certainly wasn't the first to do this.
https://www.youtube.com/c/%E6%9C%89%E9%9A%A3%E5%A0%82%E3%81%...
To be clear, they aren't just cutting out silence, they are cutting them to the point that it's a style and the speech pattern, particularly of the Owl character, sounds unnatural, which I'm assuming is what they were going for. It's that style of nearly cutting phrases together faster than they would be naturally spoken is the thing I'm saying is a trend in some Japanese videos.
Note: The channel itself is run by a stationary store chain from Tokyo. The funny part is the Owl character often makes fun of what they're showing as in "why would anyone buy this?" or "That's way too expensive" which is funny for a channel run by a store selling most of things they're showing off.
https://www.descript.com/ comes to mind
[1] https://www.youtube.com/watch?v=DQ8orIurGxw
[2] https://jumpcutter.com
[3] https://github.com/carykh/jumpcutter
With multiple speakers at once, I think removing a lot of silence can make it sound pretty unnatural.
As a solo person speaking to a camera, or recording a screencast, when I'm not perfect out the gate (read: I repeat each line like 6 times), I've found it easiest to take a breath and leave a pause between each take. After that, the editing is pretty straightforward, just cutting out the silences and keeping the good takes.
Some people are super comfortable on camera and can just talk for a few minutes and it's done... or memorize a script and knock it out... but sadly I'm not one of those people. Even with upfront planning, I'm more like "record 45 minutes to make a 5 minute screencast" so removing silence automatically is a huge time saver.
https://manual.audacityteam.org/man/truncate_silence.html
Editing audio to deal with "silent sections" is nothing like editing the video for the corresponding sections.
This approach was the one I tried first also (I also tried the frequency one fwiw, which has its own, worse drawbacks). But using loudness runs into issues if the source loudness isn't (relatively) even across the entire source media. Using a single sensitivity setting like this would be a problem if:
* recording gain is set to automatic, and there are sudden changes in noise floor like wind (if recorded in 24-bit or lower)
* crew adjusts gain partway through recording (big no-no but happens)
* talent/host moves in and out of microphone sweet spot
* talent/host adjusts themselves in a squeaky chair during silence or transition-to-silence (or coughs, or breaths loudly, or ambulance goes by...)
If you apply the edit w/ a single sensitivity and something like the above is true, it would cut in the wrong place. Unfortunately, you would have to watch the entire show, skipping to boundaries with your full attention to know that ever got a cut wrong.
and yes, I was very excited, thank you for checking it out, Van!
Vidbase is looking awesome btw! I bet it's going to be huge. It looks like you've paid an insane amount of attention to the details.
Cutting also removes ums and ahs. Newsreaders take ages to get good at this at speed, so giving people these tools makes video more accessible as a publishing medium.
1. Cross fade the audio between clips, without crossfading the video, if the sound is available past the cut point. I think its pretty much 90% of the time that I would want something like that.
2. I have clips from a gopro that had a selfie stick that made a loud clicky sound from rattling that peaks above regular audio. Some way to lower the volume on that clicky noise without having to manually go through and do it for each click.
dont know how we'll deal with the second one ...
Personally I'd recommend writing a script and rehearsing it, instead of going on the fly and having to edit out the umhs, aahs and awkward silences. This might also raise the standard for excitement above the generic "super stoked".
Having said that, I think there's probably a market for this. Good luck to OP!
I will agree that those cuts in body movement seem odd if anything when silences are removed. Maybe put a cat picture in the middle.
Can you comment on differences?