International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 05 | May 2025
p-ISSN: 2395-0072
www.irjet.net
ML POWERED PERSONALIZED HEARING AID Aditya Kulkarni1, Khushi Raval2, Nishant Gangurde3, Omkar Katkar4, Suhasini Itkar5 1234 UG Student, Department of Computer Engineering, Savitribai Phule Pune University, Pune, India
5Head, Dept of Computer Engineering, PES Modern College Of Engineering, Shivajinagar, Pune, India
---------------------------------------------------------------------***--------------------------------------------------------------------1.2 Scope
Abstract - This project focuses on individuals using hearing aids who face difficulty hearing in noisy environments by introducing devices that aim to amplify all sounds equally by reducing background noise effectively. By combining traditional signal processing with advanced machine learning to deliver intelligent noise suppression, MVDR beamforming helps to isolate sounds from specific directions thereby focusing on speaker’s voice supported by dual microphone. Voice Activity Detection aids to detect and process speech segments by further processing them into the time frequency domain. In order to further enhance the clarity by classifying and adjusting audio frames, a post processing module like Support Vector Machine is applied. It runs with low delay, making it ideal for real-time hearing aids. Its modular design fits easily into other audio systems. By combining classic beamforming with AI, it offers a smarter way to help people hear better in noisy places
The system mainly focuses on enhancing the human speech or relevant sound in noisy environments for hearing aid applications. This project includes dual microphone beamforming, noise suppression, and classification using machine learning knowledge. It does not rely on IoT integration, biometric authentication, or commercial deployment. Instead of this, it provides a software solution that can well integrate with hearing aid devices or the mobile application for being user-friendly.
2. SYSTEM ARCHITECURE The system arch is designed to enhance the real-time audio signals for hearing aid users by considering signal processing with machine learning. The process starts with acquisition of noisy audio input from the surrounding environment, through the microphone. This raw input is then first converted into the frequency domain using the STFT, i.e., the Short-Time Fourier Transform. STFT divides the audio signal into smaller overlapping windows and also applies Fourier transform to each segment frame . This generates the time-frequency representation, where each frame tells about how frequency content evolves over time. STFT allows the system to analyze speech features while also detecting and analyzing background noise, making it ideal for further enhancement stages.
Key Words: Audio Processing, MVDR, VAD, Beamforming, Noise Reduction, Machine learning, Signal-to-Noise Ratio (SNR), Hearing Assistance.
1.INTRODUCTION This project aims to provide an intelligent, ML-powered hearing aid system for improving relevant sound in a noisy environment. Our system mainly enhances speech while suppressing background noise using advanced signal processing steps.
Next, the system applies Voice Activity Detection (VAD) to differentiate between human speech and non-speech segments. This step helps to focus only on segments where speech is actually present, enhancing both accuracy and computation. Following this, the Minimum Variance Distortion less Response (MVDR) beamforming algorithm is used to spatially filter out the incoming audio signal. MVDR works by steering the beam of the microphone array towards the specific speech source while reducing the power from all other directions. This is done by calculating the beamformer weight vector that maintains the desired signal without distortion while suppressing the background interferences. MVDR dynamically adapts to the noisy environment by calculating the covariance matrix, helping to suppress moving noise sources.
The key components of this project are Short-Time Fourier Transform (STFT), Voice Activity Detection (VAD), MVDR beamforming, and SVM-based classification. Real-time audio is given as input and then processed, filtered, and reconstructed to deliver the final clear audio to the user.
1.1 Motivation Due to increasing noise pollution in the public and daily environment, it becomes difficult for individuals with hearing problems to differentiate human speech from background noise. Normal hearing aids amplify all the sounds in the surroundings, making it difficult to focus on the required and essential sounds that need enhancement. So, there is a demanding need for such a hearing aid that will amplify only the relevant sounds while reducing the background noise. The advanced technologies in Machine Learning and signal processing help to tackle these problems and provide a better and more natural hearing experience.
© 2025, IRJET
|
Impact Factor value: 8.315
The beamformed signal is then passed to the adaptive filtering stage, where a noise covariance matrix is continuously estimated. This helps to apply targeted noise reduction by adapting filter coefficients in real-time,
|
ISO 9001:2008 Certified Journal
|
Page 1638