I am the host of a German-speaking software engineering podcast (https://engineeringkiosk.dev/).
This is a side project, and publishing one episode every week is challenging. Primarily if you target an evergreen strategy and not a news podcast.<p>Things like preparing a topic, recording, audio editing, writing shownotes/links/chapters, marketing (social media posts), and maybe editing reels/shorts. This can quickly consume ~8h per week.<p>We looked into the usage of AI with the goal of reducing the time spent.<p>Right now we are using https://www.assemblyai.com/ to generate transcripts. This helps us a lot because we often ask ourselves, "Didn't we talk about X in some episodes already?" - DUe to transcripts, we can easily search it.<p>We are experimenting with https://podsqueeze.com/ to auto-generate chapters, show notes, auto-extract links etc. - In the end, this seems to be a service that generates transcripts and sends them to ChatGPT with a few custom prompts - We don't use the full service, but might end up with our own workflow doing the same. Still a good inspiration for now.<p>When recording is going wrong and we only end up with one audio track (vs. 1 audio track per speaker), we use https://auphonic.com/ to normalize the audio loudness, etc. This service is doing a pretty good job.<p>Right now, we are playing around with the automatic removal of filler words (like "ehms") in the audio file with the help of a transcript. The challenge here is that we record in German. Many AI's don't support the German language as well as the English one. Time will tell.<p>However, we can say the transcripts had the biggest improvement in our workflow - The speed up writing the show notes a lot!<p>How do you use AI to speed up / get support for your podcasting workflow?
Yes and no<p>For Cybershow [0] we love to use AI images, from a variety of
generators, because it's nice to have quick, catchy iconic pictures
that help listeners remember/search the episodes visually.<p>Also most of the main cast channels now do great transcription, and
our audio is clear, so that's one more job that's automatically done
for us these days by "AI".<p>In the production and audio domain there's very little that "AI" can
do better than an experienced writer and editor, imho as a half-decent
writer/editor :).<p>A lot of editing is moment by moment decision making that just
requires a good ear. "AI" can't tell me if something a guest says is
ambiguous, or sounds a bit poorly evidenced etc.<p>Maybe one tool I'd maybe use if it existed is a "fluff remover" for
"ummmm, like, well... y'know, ummmmm, literally... like...."<p>But I wouldn't trust it not to butcher the audio and leave things in
an irrecoverable (no undo path) state that only gets spotted in the
mastering stage.<p>Otherwise, many "intelligent" plugins already have built in clever DSP
features these days, for things like noise reduction, silence removal,
level matching.<p>Scripting things is the key... and Audacity is a dream for this!<p>For planning/organisation I use emacs org-mode which is fine for
capturing notes, linking resources and pulling it all together into a
script or talking-points list for recording.<p>For an amateur show we don't need much more quality than this.<p>[0] <a href="https://cybershow.uk/" rel="nofollow noreferrer">https://cybershow.uk/</a>