In Part 1 I found that my Malayalam speech model’s test set leaked into its training data. In Part 2 I re-tested the model on clean data only, and promised a better v2. This post reports two things. First, v2 is trained, and on the same clean test set it cuts word errors by 12
Part 3: Retraining My Malayalam ASR Model, and a Bug in My Own Evaluation Read More »
Correction (11 October 2026): The v1 numbers in this post came from an evaluation notebook that capped transcripts at 225 tokens. Malayalam needs about 25 Whisper tokens per second of speech, so clips longer than about 9 seconds were cut off and the missing words counted as errors. With the cap fixed, v1 scores 43.5%
I Re-Tested My Malayalam ASR Model on Clean Data — Part 2 Read More »
Update (October 2026): This is Part 1 of a three-part series. Part 2 re-tests v1 on the clean test set built here. Part 3 reports the finished v2 (43.5% → 31.2% WER) and a bug in my own evaluation. Three things in this post have changed since I wrote it: v2 is trained. It used
I Checked My Own ASR Dataset for Leakage — Here’s What I Found Read More »
Softmax activation function is a commonly used activation function in deep learning. In this tutorial, let us see what it is, why it is needed, and how it is used. Softmax activation function is commonly used in the output layer of classifier neural network. What is a classifier? Lets start with understanding what a classifier
Softmax Activation Function Read More »
Introduction TinyML is the field which involves deployment of machine learning models into resource constrained devices such as micro-controllers. Such devices called edge devices often have few Kilobytes of RAM and flash memory but consumes power in milli-watts range. This feature makes the technology an ideal choice for remote sensing applications, weather stations, tiny gadgets
Approximating Sine Function in TinyML Read More »
The concept of the Internet of Things (IoT) revolves around the integration of sensors with compact computing devices like Arduino boards. This fusion enables a multitude of applications, ranging from surveillance monitoring to automated door locking and smart farming. A pivotal component of IoT systems is the dashboard, which serves as a hub for visualizing
A Simple Data Visualization Web App Using Streamlit Library Read More »
User Interface is key for humans to interact with computers. The traditional way is to use keyboard and mouse to browse websites, use apps, or play video games. When it comes to specific applications such as infographics display in a mall or airport often the amount of content will be limitted and simple. In such
User Interface Control Using Hand Gesture Recognition Read More »
In many situations, it is desirable to start certain programs or softwares automatically while a computer boots up. In this article we will see few ways we can achieve this objective using some settings in Linux systems. The Bashrc method This is one of the easiest methods to load a program or script while a
Auto-Initialization Scripts in Linux Read More »
Defining neural network architecture is a key step in deep learning. A lot of decisions have to be made regarding the architecture, number of nodes, activations function, regularization etc. One of the the important aspect is combining multiple inputs parallely and passing it to the next layer. This is done using the Functional API of
Sequential API and Functional API in Keras Read More »
Hello World, In this article let us take a look at the process of matrix multiplication from a different perspective. Often matrix multiplication is taught or learned as a boring process of certain rules. This spoils the fun of doing it. So why don’t we look for better way to think about it by having
Matrix Multiplication in Real-Life Read More »