Although I'm not one of the RSS absolutists who think no RSS = not a podcast (I think language changes over time, for example for me any vacuum cleaner is a Hoover and any temporary building a Portakabin) I do totally agree with them about the power of RSS.
Having an open standard allows innovation from a wide range of people and as I'll discuss below allows automation and the ability to use the information held in the RSS in a variety of ways (also see Podcast Index for how RSS is being used in podcast app development).
The aim of this piece is to show what can be done with little or no previous knowledge and also a call for anyone who's interested in these things to get in touch, it's a bit rough around the edges at the moment but I'd rather get some (or no) feedback now rather than spending a month polishing something nobody is interested in.
Historically any programming required a certain level of knowledge and faffing about downloading the correct things, setting up in a specific way etc., that for a lot of people is more trouble than its worth and totally out of their comfort zone.
The rise in cloud computing and open source however means that a kid in their bedroom can have access to the same tools (with maybe a few size/speed restrictions) that people at a trillion dollar company use and also have the ability to share what they create easily with anyone in the world.
The below are some basic examples of the kind of things that can be done with Google Colab and RSS, if only some of the words in the image below make sense to you that doesn't matter, the aim of this is to make it as easy as possible for a user (other tools that do similar things to Colab are available such as Posit Cloud).
This was inspired in part by a Tweet by previous podcast guest Stuart Goldsmith:
Someone had already messaged Stu to help him with this but it got me thinking about making the most of RSS.
Armed with Google Colab and Chat GPT and Stack Overflow for a bit of support, I put together some code where you put in the RSS feed at the start (can be yours, mine or anyone else's) and it'll create a csv file in your google drive with all the summary details with one row per episode.
The next thing I built was to effectively take a back up of all audio files hosted on a podcast, if you're organised maybe you already have these locally and neatly filed but if you're more like me there'll be here and there and maybe nowhere local if you've been podcasting for any length of time.
This code effectively downloads every episode to a folder and renames the files based on the podcast name and series/episode number as again if you're anything like me those files will be called something like Stu Final v2.mp3 or may get renamed to something completely different by your host.
The use of transcripts in podcasting has been something for quite some time within the open Podcasting community but Apple starting to transcribe podcasts could help bring it more to the mainstream as for accessibility it should be seen more as a necessity rather than a luxury.
If you produce a podcast there's a fair chance you use a tool such as Descript to edit it and within that have the option to output transcripts for new episodes though if you're someone such as Stu with his 400+ episodes, you wouldn't want to be manually loading each one to Descript, transcribing and outputting the results.
This is where Google Colab comes in, if we take the MP3 files we've downloaded these can then be transcribed for free using Open AIs Whisper tool - again I'd be interested to know how many people want more detail and how many just want to put their RSS details in and pop out the results at the other end.
Below is a brief example of differing levels of transcription for a trailer for the new podcast by comedians Frankie Boyle, Susie McCabe and Christopher Macarthur-Boyd. This is a good thing to test on as it's under a minute, has people with fairly strong accents and as a trailer is tightly edited so few gaps between speech than normal.
Actual Text: I'm Frankie Boyle. I'm Susie McCabe. And I'm Christopher Macarthur Boyd. And this is Here Comes the Guillotine. There is zero theme. Zero structure. Quite grim chat about the history of the planet and its future, which is death. I wouldn’t throw the words grim or death in the promo.
For a lot of the time, the studio is bathed with flame and we recreate the epic of Gilgamesh regularly. We also have a wee bit of a laugh. Sometimes I turn into a flaming ox and take to the skies. Sometimes they get a bit carried away with their goblin chat. But listen, all in all tune in. It’ll be a good laugh. It's just three pals having a chat.
Here comes the Guillotine, exclusive to global player from February
Running that through Descript would give you:
Descript: I'm Frankie Boyle. I'm Susie McCabe. And I'm Christopher Macarthur-Boyd. And this is Here Comes the Guillotine. There is zero theme. Zero structure. Quite grim chat about the history of the planet and its future, which is death. I went through the words grandma death, uh, in the promo for a lot of the time.
The studio is bathed with flame and we recreate the epic of Gilgamesh regularly. We also have a wee bit, a laugh. Sometimes I turn into a flaming ox and tape to disguise. Sometimes they get a bit carried away with our gold blue goblin chat. But listen, all in, all tune in. Beggar laugh. It's just three pals having a chat.
Here comes the Galatin, exclusive to global player from February.
Whisper from OpenAI have various 'strengths' of model that it can use to transcribe and there's a trade off between speed and accuracy and for the 46 second trailer using Google Colab it takes:
Model Speed (Seconds)
Tiny 7
Base 14
Small 37
Medium 120
As text that looks like:
Tiny Model - No Prompt: I'm Frankie Boyle. I'm Susie McKeep and I'm Christopher McArthur Boyle and this is here comes the Guilty. There is zero theme, zero structure, quite grim chat about the history of the planet and its future which is death. I mean through the words, grandma death in the promo.
For a lot of the time the studio is bathed with flame and we recreate the epic of Gilgamesh regularly. We also have a wee bit of a laugh. Sometimes I turn into a flame and auction take to this guys.
Sometimes they get a bit caddy to be with a gold blancher but less than all an old chin and a bigger glass. It's just three pounds having a chat. Here comes the Guilty, exclusive to Global Player from February.
You can however also provide a guide to the transcription as part of the process e.g., this is a podcast by x about y and detail and particularly important spellings. When this was done and the small model used the transcription was fully accurate (this won't always be the case and in my test the medium model was slightly less accurate than the small).
Ultimately what you want the output for and the amount of time you're willing the process to take will determine what this may be useful to you for. If you've 100 1 hour episodes and you just want a rough transcription for each to use to repurpose, then a tiny model running for an afternoon might be good enough and it'd be up to you if you wanted 100 separate files or 1 giant document that you could then refer to.
Transcribing is just one example of what can be done by making the most of RSS and Python but all sorts of possibilities are out there including:
Translation
Text to Speech
Editing/Amalgamating Files
Episode summaries and answering questions based on podcast content
Ultimately RSS gives an openness and flexibility that enables all sorts of creativity to be layered over the top and subsequent articles will look to develop this further.
Any questions/requests get in touch and if you want to be a guinea pig for some of the features mentioned above, drop me a message with your RSS.
Website: https://podcasting.makingcreativitypay.com/episodes/
Twitter: https://twitter.com/Beyond_the_Pod
