PodSearch Shows That AI Can Be Awesome, and Not Just Evil—Here's How

Artificial intelligence (AI) tech can be useful for way more than bringing about the copyright apocalypse or letting kids cheat on exams.

PodSearch uses OpenAI's Whisper speech-to-text tool to turn podcast episodes into a readable, searchable text. Right now, there are just a handful of shows available, but there's no reason that a podcast directory couldn't do this for all podcasts, and you could even–in theory–run an app on your own computer to do it for you. This is a big deal, not just for podcast listeners and researchers, but for AI as a general tool instead of as a gimmick.

"This kind of resource would be invaluable for researchers who need to review multiple podcast recordings to gather data on a certain topic. The ability to search and filter through podcast transcripts would make this process far more streamlined and efficient," Oliver Goodwin, founder and CEO of AI audio and video platform Synthesys told Lifewire via email.

OpenAI's Whisper

Frankly, Whisper is incredible. On the project page at OpenAI (the company behind Dall-E and ChatGPT), you can check out some of the examples. One is a clip of the old Micro Machines ad, with a low-quality recording of a man speaking very fast. I can hardly make out what he's saying, yet the transcript looks pretty much perfect.

This ability to pull words out of confusing, noisy contexts makes it perfect for transcribing podcasts. David Smith, the developer of PodSearch, and the creator of Pedometer++ and the smash-hit Watchsmith, first built PodSearch in 2017, but the results were less than satisfactory. Now, he has updated it, using Whisper to do the transcription.

"In the original version of this the transcription was done with very rough, slow, low accuracy system, which resulted in transcripts that were fit for keyword searching but not much else," Smith said on his blog. 

He says that Whisper "results in transcripts that are shockingly good, almost as good as a human transcriber." When you check out the comparisons between old and new on the PodSearch site, you'll see just how much better they are.

The Value of Podcast Transcripts

Podcasts are somewhat impenetrable. After you've listened to them, they're mostly useless. Any Google searches only have the title and taglines to go on and show notes from some of the more forward-thinking podcast creators.  This is a shame because they are full of all kinds of information.

Imagine if all podcasts had accurate transcriptions, right down to technical terminology, odd brand names, and so on. You'd immediately be able to search on a huge trove of information. It would be great for researchers, journalists, or anyone. And accurate speech-to-text isn't just for podcasts. Anytime you need to get spoken words into text accurately, OpenAI's Whisper could be the answer. 

"AI tech could also be used to recognize and transcribe spoken words more accurately and quickly, creating a valuable resource for those with disabilities."

How about YouTube videos? They have automatic subtitles, but they're usually pretty bad. And while we're about it, YouTube transcriptions are also a valuable search resource. You could also combine Whisper with other AI apps. Instead of using tricky dictation software, for example, you could just record yourself speaking naturally, use Whisper to transcribe the recording, then use something like ChatGPT to clean up and summarize the result. 

Whisper is not an app you can use yourself. It's a developer resource right now. But if you want to try it out, developer Jordi Bruin has used it to create Whisper Transcription, a Mac app for, you guessed it, transcription.

AI Isn't All Bad

But not everyone will use OpenAI's Whisper for good. It seems that no matter what technology invents, the marketers will soon use it to sell things to us more efficiently.

"This technology could be used to create detailed analytics on podcasts," says Goodwin. "In addition, AI could be used to detect patterns and emotions in actual conversations, providing valuable insights for businesses and marketers."

Fortunately, marketers don't have much access to our speech—yet. And until then, we can enjoy a genuinely useful AI tool that will make a lot of things better and easier. We've gotten used to some of the negative aspects of AI–the copyright problems, for example—but when it also enables tools like Whisper, we start to see how AI can benefit us and not just make things worse.