MERIN CYRIAC

Context & Motivation

Many speech-based systems rely on cloud processing, which limits their usefulness in assistive applications and low-connectivity environments. This project explored whether simple speech-driven interaction could be implemented entirely offline on constrained hardware, using non-visual feedback suitable for assistive contexts.

Problem Statement

Design and implement a system that:

• Detects predefined spoken keywords without internet access
• Operates on low-power, resource-limited hardware
• Provides reliable feedback without using a display

System Overview

The system consists of a microphone for audio input, a Raspberry Pi for offline processing, and a vibration motor for tactile feedback. The design intentionally focused on small keyword sets rather than large-vocabulary recognition, prioritising responsiveness and reliability under limited resources.

Methodology

Audio Data Handling

• Collected small audio samples for predefined keywords
• Worked with limited datasets to reflect realistic deployment constraints
• Observed variability in speech patterns and environmental noise

Offline Recognition Approach

• Explored lightweight offline speech recognition techniques suitable for Raspberry Pi
• Selected approaches that balanced computational cost and responsiveness
• Evaluated false activations and missed detections during testing

Hardware Integration

• Interfaced microphone input with Raspberry Pi
• Controlled vibration motor using GPIO
• Ensured stable motor activation without disrupting audio processing

Challenges

• Reduced recognition reliability due to small datasets
• Sensitivity to background noise
• Trade-offs between processing speed and detection accuracy
• Debugging timing issues between software processing and hardware output

Results

The system successfully detected predefined keywords entirely offline and triggered tactile feedback with acceptable latency. While limited in scope, the prototype demonstrated the feasibility of basic speech-driven interaction without network dependency.

Learnings

• Signal quality influenced system performance more than algorithm complexity
• Clearly defined constraints simplified design decisions
• Tactile feedback reduced dependence on visual interfaces
• Reliability issues often originated at the signal acquisition stage

Skills & Tools

Hardware & Electronics

• Circuit analysis and basic signal conditioning
• Sensor interfacing and prototyping
• Hardware testing and debugging

Embedded & Software

• Raspberry Pi and Arduino
• Python for scripting and prototyping
• GPIO-based hardware control

Systems & Tooling

• Linux-based development environments
• Git and version control

Interests & Focus Areas

• Assistive Technology
• Digital Fabrication
• DIY Electronics Projects
• Technical Writing