Posts

Showing posts with the label making-vaas

Speech-to-text

Image
I've started the main stretch in the implementation of VAaS: the actual coding. Currently, I have done a few different things: Speech-to-text I've started with speech-to-text for emotion recognition. Initially, I wanted to have an AI analyze the emotion of the speaker from the direct audio file, and while that still isn't impossible, it is leagues easier to train an AI like BERT to analyze the emotion in the text itself. But I'm getting ahead of myself here. So far, I've used a simple python library to transcribe a couple of songs' lyrics. You can follow along with me, since I'm planning to make everything I learn into a tutorial. The first speech-to-text tutorial is here (on Google Colab). Simple Sentiment Analysis Using BERT Honestly, I'm still not totally sure how this code works. But, I have taken the pretrained BERT model and run a sentiment analysis on two simple sentences. The goal of the notebook was to identify if those sentences conveyed a pos...

Finding Training Data for VAaS (7/5)

I've been struggling to find data to train VAaS on. Though I wanted to start small - train a neural net such as TensorFlow ASR on vocal recordings of users who are drunk. However, every single corpora that I could find online has a restriction where I need an advisor at a university. Here are the ones I found but couldn't get my hands on: Alcohol Language Corpus DementiaBank PsychosisBank There are many more that are there on the sites linked, but they all require some sort of university login, whether through CLARIN or through a CMU login. Unfortunately, I have nether, but I have emailed the professors who are in charge of login information I doubt I will get a response, though, since I asked them in the dead of summer.

Who is VAaS?

Image
The purpose of this blog is to detail the creation of VAaS, an audio-based neural net whose purpose is to detect and stop drug-based overdoses purely based on vocal data. VAaS is my (Niko Kompella's) passion project. In this drug, I hope to also provide information on drug research in the San Francisco Bay Area. At the end, I will publish a guide on another site (prototype here ) for parents and educators in the SF Bay Area detailing guides on the different drugs a high schooler may encounter, as well as preventative measures for addiction. Why VAaS? VAaS is a name with two meanings - an acronym, and an allegory. VAaS, firstly, stands for V ocal A nalysis for a buse of S ubstance. The acronym captures everything the project aims to do pretty concisely. VAaS is also named after Vaas Montenegro, the acclaimed antagonist of the game Far Cry 3. I chose Vaas for two reasons.  First of all, he's a pretty accurate depiction of the horrors of drug abuse - he literally heads a group of ...