Integrating Assembly AI with Airtable for Efficient Transcription (Full Transcript)

Speaker 1: Today, I'll be demonstrating how I integrated Assembly AI Speech API directly into my Airtable database so that I can generate and manage transcripts at scale. So by directly utilizing an API, it's basically the cost-efficient or wholesale way of accessing AI features instead of going through a third-party service. And Airtable as a relational database allows us to build our own apps, which integrate with a variety of data sources, including APIs. So we'll hop into Airtable here, and what I have is a conversation with a theoretical physicist, and I already have my audio file uploaded, and I have a public URL that Assembly AI API will need to access the file. I'm expecting two speakers to be in here, and now I will post the transcript to the API or post the transcript data, public URL, and some settings. And we're going to see here that in automations running in the background is a script making that call and writing this data back. So transcript ID, we have the status queued. I check my Assembly AI. Here, it's processing. I can see the file in the queue. And what we will get in a few moments here is a webhook will talk to my Airtable database, and it just did right there. So it's transcript get is another function. It's another script which is going to retrieve the transcript and format it as a readable text, which is what we just got in here. So here's the transcript. We're going to take a closer look at that and play the audio file along with it. There's also some other interesting metadata from the API. This file is only 10 minutes, so you can get transcript chapters as well, but we only have one chapter. Let me show you this real quick. So we'll change the speaker names that are assigned here. So interviewer and speaker B will be physicist. And if I commit the speakers... Okay, so that just ran through and updated speakers. I have one automation that can take this to Google Drive, take that text and give us a Google Drive file. I also have an integration here with document.me to generate a PDF. So I'm going to do that. And as this is generating, I'll pull up the audio file and we'll listen along with it to see how accurate this is. The PDF is generated. And let me jump in to show you quick this document.me service where you can set up these variables. And then this will talk to your AirTable base in order to allow you to basically send the data from the base and actually make a PDF out of it. So it works really great. And now we'll play the audio file and just compare a little bit of this text.

Speaker 2: When did you first come up with dilation theory?

Speaker 3: I think the first inklings of it started when I was a boy, when I was a teenager, about 13 years old, after I watched something quite fantastic in the skies above my family home.

Speaker 2: What did you actually...

Speaker 1: We'll stop it there. But yeah, that's accurate so far. That looked 100% to me. And the rest of it is probably really good too. I've used this API quite a bit. And it's definitely more detailed than some of the other APIs. I was originally using OpenAI's speech-to-text, the Whisper model that they offer through the API, but the output was not as detailed as AssemblyAI's. So I ended up going with AssemblyAI. Okay, if you like this demo and could benefit from bulk transcription services, please follow my links and I can get this set up in your workflow. Thank you.