# How to Prepare a Dataset for Hume's Custom Models ## Metadata - **Published:** 8/26/2024 - **Duration:** 7 minutes - **YouTube URL:** https://youtube.com/watch?v=PP_EhGQfZbs - **Channel:** nerding.io ## Description Unlock the power of predictive AI with Hume’s Custom Model API! Now you can predict well-being, satisfaction, mental health, and more by creating custom multimodal models using just a few labeled examples. Key Highlights: - Integrate dynamic patterns of language, vocal, and facial expressions into your models - Leverage Hume’s pretrained AI, fine-tuned on millions of videos and audio files - Achieve accurate predictions with just a few dozen examples - Deploy and share models tailored specifically to your needs Start predicting the outcomes that matter most to your users with minimal effort. Like, comment, and subscribe for more AI insights! 📰 News & Resources: https://sendfox.com/nerdingio 📞 Book a Call: https://calendar.app.google/M1iU6X2x18metzDeA 🎥 Chapters 00:00 Introduction 🔗 Links https://www.hume.ai/ https://www.hume.ai/blog/hume-custom-model-api ⤵️ Let's Connect https://everefficient.ai https://nerding.io https://twitter.com/nerding_io https://www.linkedin.com/in/jdfiscus/ https://www.linkedin.com/company/ever-efficient-ai/ #chatgpt #voice #ai #programming ## Key Highlights ### 1. Hume AI Custom Model Overview The video introduces Hume AI and its custom model building capabilities, covering exploration of the playground and understanding different model types offered by the platform. ### 2. Playground Exploration & Key Features The video demonstrates Hume AI's playground features including expression measurement via webcam, voice burst detection, and speech prosody analysis with embedding plots, showcasing real-time emotion and speech analysis. ### 3. Custom Model Dataset Preparation Explains how to prepare a dataset for custom models by organizing image/video/audio files into labeled subfolders (e.g., 'attentive' and 'distracted') within a root directory, then uploading it to Hume AI. ### 4. Building Custom Models: A Simple Process The process to create custom models is relatively straightforward within Hume AI's platform. Getting and labeling the data correctly, then uploading, seems to be the most challenging step. ## Summary ## Hume AI Custom Model Creation: Video Summary **1. Executive Summary:** This video provides an overview of Hume AI's custom model building capabilities, focusing on preparing a dataset for creating personalized AI models that can predict traits like well-being, satisfaction, or mental health from multimodal data (audio, video, text). It highlights the ease of use of the Hume AI platform and how to prepare data for custom model creation. **2. Main Topics Covered:** * **Hume AI Custom Model Overview:** Introduction to Hume AI and its custom model building features. * **Playground Exploration & Key Features:** Demonstration of Hume AI's playground features including expression measurement via webcam, voice burst detection, and speech prosody analysis with embedding plots. This section showcases real-time emotion and speech analysis. * **Custom Model Dataset Preparation:** A detailed explanation of how to structure and prepare a dataset for custom models. * **Building Custom Models:** A process to create custom models within Hume AI's platform. Getting and labeling the data correctly, then uploading, are key steps. **3. Key Takeaways:** * Hume AI allows users to build custom AI models with minimal effort by leveraging pre-trained AI fine-tuned on millions of videos and audio files. * Custom models can predict a variety of outcomes related to well-being, mental health, and satisfaction using multimodal data. * Preparing a dataset involves organizing image, video, or audio files into labeled subfolders (e.g., 'attentive' and 'distracted') within a root directory. * Creating a custom model is a relatively straightforward process on the Hume AI platform once the data is correctly structured and labeled. * The most challenging aspect is data acquisition and accurate labeling of that data. **4. Notable Quotes or Examples:** * "Sort your image your video and your audio files into subfolders based on their labels..." - Example of dataset preparation instructions from Hume AI. * The video showcases examples of Hume AI's playground functionality, including real-time facial expression analysis and voice burst detection, highlighting the platform's capabilities. * An example is given of creating a model that detects "attentive" vs. "distracted" behavior. **5. Target Audience:** * AI Developers and Engineers * Data Scientists interested in Multimodal AI * Researchers exploring AI-driven insights into human behavior * Anyone interested in building custom AI models for specific prediction tasks. ## Full Transcript hey everyone welcome to nerding IO I'm JD and today what we're going to be talking about is Hume Ai and how to do custom models and so this means we're going to look at the process of how you can actually build your own model we're also going to get through the playground and understand the different types of models they have and with that let's go ahead and get started hey everyone what we're going to be doing is going through Hume and first we're going to start with their playground so the first thing we want to do is log in we're going to go to the expression measurement playground and just kind of see like some of the uh the models that we can actually interact with and stuff like that so if we start our webcam we can kind of see that it is giving us different facial expressions I'm getting confusion and concentration a lot which is entertaining and then as we're speaking we can actually see what's going on with this embedding plot and this is pretty interesting because then we can kind of relate to the all the different quadrants of this uh this kind of embedding so I just wanted to show that as like something pretty cool but they also have this ability to do a voice burst so as you're speaking you can see different types of detection and frustration and determination and also in this uh in this voice plot the next one is the speech parody and so or paracity and we've kind of seen this before right where we're seeing it in our uh applications that we've tried so far but again it has the the embedding plot real quick everyone if you haven't already please remember to like And subscribe it helps more than you know with that let's get back to it so the other thing that we can do is actually select a model from one of our uh ex like from one of the models that they have right they have image models Audio models video models and then obviously text models we can either try an example upload files or create our own models and so what we're going to do is just kind of look at this uh as we're going to go through this attentive and distractive one so you can just do like a try an example and select a click clip and then it will go through and actually analyze that file so we're going to come back to this later on and actually just look at the guts of how to put this together the other thing is the uh text editor which which we kind of know so now we're going to actually take a look at the custom models and as you could see through the examples in the playground there's a bunch of different ones and so we're going to be kind of following this path of this attentive and distractive just distract Ed and if you click on this you can kind of see there's all the the different quadrants is labeled right if it's distracted or kind of in the middle then you can actually see the predictions as well as the labels you can actually pull up the videos of this data set that they've looked at and you can actually click on the data set itself so so if we actually go to this data set you can kind of see that there's not like a ton of data right in this specific example obviously the more data you have the better you can see it's kind of split into two areas of focus there is distracted and there's about 60 videos I think in here of data to be analyzed and looked at and they kind of split it in between and so it's actually um pretty simple as to how you build your your models and so they they have this tutorial where you can actually go through and take your own data set now I don't have a ton of data um to actually create this so we're just going to go through the example of how you can do this so if you look at the right here it says sort your image your video and your audio files into subfolders based on their labels and this video is like super simple I'm just going to kind of describe what's happening so basically you can start with a root folder and that's going to be your label of student Focus then you're putting all your distracted in and you can see that or Focus or distracted and you can see the labels right you're just changing the file name and then underscore 01 or underscore 02 these are just your examples and you're putting all of this in a root then what it's doing is you're saying okay now that you have all these you need subfolders to actually these are the uh the labels themselves or the annotations that you're you're putting in there and you have attentative and distractive and so now you can take this and actually upload it into a uh data set so all you're really going to need to do is go and go to the the models you can create a custom data set and then that is where you're going to be following like what kind of columns all of them would be car uh categorical and then you would go ahead and upload all of that model and then all you have to do is really run it and you can analyze it so this tutorial is actually super helpful it's it's pretty straightforward the videos are are pretty quick so if you actually want to go through and build your own model it's a pretty easy step-by-step basis the hardest part really is getting all of your data and then making sure that it's labeled correctly and then uploading it so again if you were to think about the how we would separate this as a folder you would have your distracted folder you would have your attentative folder and then you would just have maybe even the folder called attentative verse distracted and that's all you really need to do in order to get started in building your own custom models which you can then use in uh in the uh playground all right that's it for us today everyone what we went through was humi and the ability to look at different types of custom models and how you could actually create a data set in order to make your own model and interact with it with that happy nerdy --- *Generated for LLM consumption from nerding.io video library*