A Multi Module Deep Learning Framework for User Preferences and Behaviour in FM Radio

View at publisher Summarise with VIRAT
Authors: C.Selvarathi, G.Naveen Kumar, A.Kumar, P. Pandiaraja, Charles Prabu V, S.Maheswari
Cited by: 0
References: 15

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The Speech Processing Project is a broad process of giving meaning to the audio with the help of state-of-the-art machine learning and deep learning. This project brings together a number of key modules in order to reach some very important goals, which include preparing audio, speech transcription, inappropriate content detection, identifying song preferences, and estimating the age of speakers. Taking advantage of state-of-the-art models like OpenAI's Whisper for converting speech to text, Facebook's Wav2Vec2 for classifying audio, and Cardiff NLP for detecting offensive material, the following project offers a powerful yet flexible means to analyze a wide variety of audio information. The preprocessing pipeline employs tools such as Librosa and Noisereduce for noise reduction, normalization, and dynamic range compression to make sure that high-quality audio goes into the models. Besides, in-house machine learning models with Scikit-learn-trained algorithms attained a mean squared error (MSE) of 0.35 for speaker age estimation and more than 95% accuracy for transcription analysis. The module for song preference analysis was seen to be able to classify songs by era with 90% confidence and map preferences to age groups successfully. Through automating music and conversation audio analysis, the project provides outputs as accurate transcripts, confidently detected inappropriate content, song tastes by time ranges and age groups, and results of age estimation. Provided through interactive visualizations and reports in details, the project outputs indicate that it can handle real-world tasks for audio analysis and promoting precision and scalability across applications.

Fields of Study & Subjects

Music and Audio Processing Speech Recognition and Synthesis Speech and Audio Processing

Publication details

Journal / source
International Conference on Image Processing and Capsule Networks (ICIPCN)
Pages
1069–1074
Publication date
1 January 2026
Language
EN