Implementation of Voice Recognition Technology through TinyML

Session Number

2

Advisor(s)

Dr. Phadmakar Patankar, IMSA

Location

A129

Discipline

Engineering

Start Date

15-4-2026 2:15 PM

End Date

15-4-2026 3:00 PM

Abstract

This project focuses on the implementation of machine learning-based voice recognition on a resource limited microcontroller platform. The primary goal is to design and integrate a system capable of identifying five distinct languages - English, Spanish, French, Arabic, and Mandarin - using real-time spoken input. The system captures user speech through a voice recognition sensor, converts the audio into text, and classifies the detected language. Hardware focuses on an Arduino-based microcontroller combined with Tiny Machine Learning (TinyML) techniques to use efficient processing. Recent advances in deep learning and feature extraction methods have improved speech recognition accuracy, and TinyML has made it possible to use these models on low-power edge devices. To build on this, the project uses Edge Impulse along with libraries from Mozilla to evaluate whether reliable multilingual recognition can be attained even with processing constraints. The final system will assess the feasibility of real-time language identification on embedded hardware and explore more potential applications of speech-based interfaces on edge devices.

Share

COinS
 
Apr 15th, 2:15 PM Apr 15th, 3:00 PM

Implementation of Voice Recognition Technology through TinyML

A129

This project focuses on the implementation of machine learning-based voice recognition on a resource limited microcontroller platform. The primary goal is to design and integrate a system capable of identifying five distinct languages - English, Spanish, French, Arabic, and Mandarin - using real-time spoken input. The system captures user speech through a voice recognition sensor, converts the audio into text, and classifies the detected language. Hardware focuses on an Arduino-based microcontroller combined with Tiny Machine Learning (TinyML) techniques to use efficient processing. Recent advances in deep learning and feature extraction methods have improved speech recognition accuracy, and TinyML has made it possible to use these models on low-power edge devices. To build on this, the project uses Edge Impulse along with libraries from Mozilla to evaluate whether reliable multilingual recognition can be attained even with processing constraints. The final system will assess the feasibility of real-time language identification on embedded hardware and explore more potential applications of speech-based interfaces on edge devices.