Dept uses Descript for some of their podcasting. I'm using it like a maniac, I think they're surprised at how into it I am. Check out the transcript & see how it performed.

• Descript
• The Ship It Podcast How to ship software, from the front lines. We talk with software developers about their craft, developer tools, developer productivity and what makes software development awesome. Hosted by your friends at Rocket Insights. AKA shipit.io
• Brandbeats Podcast by BASIC An agency podcast with views on design, technology, art, and culture. Explore the new microsite at www.brandbeats.basicagency.com

[00:00:00] Welcome back to machine learning applied. We're not going to discuss anything technical in this episode. Just a little bit of fun. I want to show you a cool project I'm using, The project is called Descript.

[00:00:09] It's a podcast recording tool software on your Mac or windows. And I'm using it to record this episode. And I used it to record the last few episodes ever since I came on to Dept because Dept's podcasting branch has a pro subscription to their service. And I am just over the moon about this product

[00:00:27] And it's just so fun and it's inspirational. Maybe it'll get you noodling on what you want to build for your next machine learning project. Just kind of show you the capabilities of machine learning out in the wild.

[00:00:37] So the first feature it offers is that it generates transcripts of your podcast episodes. So that is speech to text nothing special, but that's kind of cool. That's really valuable for me. Now I've been recording my episodes, generating a transcript and then posting them on my website show notes.

[00:00:52] So now I have transcripts that's pretty valuable for accessibility, or if somebody wants to see where in an episode, some topic I'm discussing so they can skip to that section. The next thing is that you can edit the transcript. And so if I highlight a chunk of text and click delete, it wi ll delete that from the audio, from the recording, not just from the transcript.

[00:01:13] That's pretty cool. So it reflects back and then take it to the next level. it's not only limited to. Deleting text from your transcripts. You can add text to your transcript and it will sort of use deep fake to use your voice, to generate audio back into your podcast file.

[00:01:34] So this whole section here, I didn't say any of this. This is all generated using Descript. Maybe not the best sounding Tyler impression, but really valuable if I make a one or two word error. For example in the last episode, I kept saying Client VPC when I should have said Client VPN.

[00:01:52] now of course, that didn't sound perfect. I'm sure I need to sit down a bit with the way that you submit your voice training data and get that all

[00:01:59] another feature, is it automatically removes filler words for you? So, um, and, uh, and like the other day, so like man, whatever, and you know what I mean? And, Well, I don't know. I was just hanging out and I don't know this guy came up to me or whatever. So let me just replay that back to you with the filler words removed.

[00:02:19] another feature, is it automatically removes filler words for you? and, and the other day, so man, and And, I don't know. I was just hanging out and I don't know this guy came up to me And then my favorite feature of all is that it will take your audio and it will run it through an AI filter called sound studio, And that will improve the quality of your recording using AI. So previously, what I would do is record my episodes in audacity, and I would go through a sequence of steps for editing those episodes to sound better.

[00:02:55] So the first would be to generate a noise profile, and then I'd highlight like five seconds of noise and then apply noise reduction based off of that chunk against the whole recording.

[00:03:06] Then I would apply this effect called a compressor that would sort of normalize the silent parts of the audio to accommodate against the loud parts of the audio.

[00:03:15] Then I would normalize the audio to a certain decibels range

[00:03:19] and then export the file as mono, And this did a pretty good job of cleaning up my audio, But because I don't have really high grade professional sound equipment it still wasn't the greatest. And so now I can just use my Mac mic, just my air pods, or, or my computer's built in microphone to record with Apply the sound studio filter, and rather than going through a series of applying effects, Descript will use AI to alter the audio to sound like it was professionally recorded based on, I presume some training data coming from common, professionally recorded podcasts, like the Joe Rogan show or something.

[00:03:58] so this part was recorded on my Mac microphone. And you can tell there's just terrible quality. I can even just see as I'm recording this, the spectrograms the audiograms showing high decibel of noise, background noise, and echo in the room. then I'm going to show you this exact same thing.

[00:04:17] Replayed through sound studio.

[00:04:19] so this part was recorded on my Mac microphone. And you can tell there's just terrible quality. I can even just see as I'm recording this, the spectrograms the audiograms showing high decibel of noise, background noise, and echo in the room. then I'm going to show you this exact same thing.

[00:04:36] Replayed through sound studio.

[00:04:39] Anyway, there's no goal to this episode. I just wanted to show you some cool application of machine learning. We got text-to-speech, speech to text And a handful of other audio and text manipulation technology using some pretty clever machine learning.

[00:04:51] If you ask me

[00:04:53] and Dept is using descript to record a bunch of their episodes. There's other podcasts that are part of Dept. I'll shout out. Two of them here. One is called the ship it podcast, ship it.io by rocket insights. They're a subsidiary of depth And the description of this podcast is how to ship software from the frontlines. We'll talk with software developers about their craft developer tools, developer productivity, and what makes software development awesome.

[00:05:18] Hosted by your friends at rocket insights, AKA shipit.io.

[00:05:22] and the other depth podcast is called brand beats by basic basic is another subsidiary of Dept. And the podcast description is an agency podcast with views on design technology, art, and culture. so if you're interested in topics of shipping software, go check out the ship at IO podcast, I'll link in the show notes.

[00:05:41] And if you're interested in Design culture and art interfacing with technology. Go check out the brand beats podcast, but they're using Descript and we'll see, maybe I'll get the descript guys on this podcast so I can discuss their technology, what models they're using and such.

[00:05:55] It's really interesting to me, cause it's kind of the inverse of what I'm doing. I'm teaching you AI by way of a podcast and I'm using an AI podcast recording studio. so it's the upside down version of what I'm doing for you here

[00:06:09] If there are any other cool machine learning projects that y'all are acquainted with, you should drop a link in the show notes to this episode in the comment section.

[00:06:17] so that others can check out

[00:06:18] cool applications of AI in the wild.