SOS calling screenshot

VoiceMatchApp

"Automating wake word detection and speaker verification for rapid, secure emergency responses."

Inspiration

In emergency situations, every second counts. We wanted to create a hands-free, voice-activated solution that could not only recognize a specific wake word but also verify the speaker to ensure secure and intentional SOS triggers.

What it Does

Continuous Listening: Listens in the background for a predetermined wake word.
Speaker Verification: Confirms whether the person speaking the wake word is the same user who initially set it up.
SOS Trigger: If both the wake word and speaker match are detected, the app immediately launches an SOS activity.

How We Built It

Android + Kotlin: For developing the mobile application’s core functionalities and UI.
Speech Recognition APIs: Leveraged Android’s built-in speech recognition to continually listen and detect partial results.
OkHttp: Used for making HTTP requests to our backend server.
Custom Backend APIs + TensorFlow:
- Our backend handles audio uploads, transcription, and speaker verification.
- We use TensorFlow to power the deep learning-based speaker recognition model, which analyzes the voiceprint and confirms identity.
Android Studio: Primary IDE for coding, debugging, and building the APK.

Challenges We Ran Into

Continuous Listening: Ensuring the app transitions between listening sessions without crashing or missing input.
Accurate Speaker Verification: Managing variations in audio input (background noise, microphone quality) while training and using TensorFlow models.
API Integration: Handling network timeouts, JSON parsing, and robust error handling when server downtime occurs.

Accomplishments That We're Proud Of

TensorFlow Integration: Successfully integrating a speaker recognition model into the backend for high-accuracy verification.
Seamless Wake Word Detection: Robust continuous listening with minimal false positives.
Responsive UI: Clear instructions, minimal clutter, and smooth user experience.

What We Learned

Audio Processing & Machine Learning: Gained insights into how to train and deploy TensorFlow models for speaker recognition.
Threading & Handlers: Balanced background tasks with real-time UI updates on Android.
Network Reliability: Understood the importance of handling various network error states, including retries and graceful failures.

What's Next for Voice Match SOS

Offline Mode: Integrate a local TensorFlow Lite model for wake word detection and speaker recognition when network connectivity is poor.
Multi-User Profiles: Allow multiple authorized voices (e.g., family members) with distinct wake words.
Expanded Emergency Features: Incorporate location sharing, automated text messages, or calls upon SOS activation.

Built With

Kotlin – Primary language for Android development
Android Studio – IDE for building and debugging the app
OkHttp – HTTP & HTTP/2 client for network operations
Android Speech Recognizer API – For handling voice input and recognition
TensorFlow – For deep learning-based speaker recognition
Backend APIs (Custom) – For handling file uploads, transcription, and SOS triggers
(Add any additional frameworks/services, e.g., cloud hosting or databases, if applicable.)

Video Presentation

https://drive.google.com/file/d/18N7QddLV6Xe9rdEa2YICAF1AP6KfNQt0/view?ts=678c7720

Project Information

Title

VoiceMatchApp

Summary

VoiceMatchApp is an Android application designed to seamlessly listen for a specific wake word, verify the speaker using a TensorFlow-based model, and trigger an SOS activity if the match is successful. This enables a secure and rapid response system for emergency situations.