Abstract

Personalized advertising is now a necessity in the dynamic environment of digital media in order to enhance user interest and marketing performance. The paper introduces a deep learning-based system that can be used to estimate age and gender using human voice to scale up demographic targeting. As the number of audio-based platforms (e.g., podcasts, streaming services, voice assistants, etc.) is quickly growing, voice data has become a highly valuable source of demographic data. Nevertheless, conventional signal-processing methods tend to lose the small-scale acoustic variation that is necessary to make the demographic inference. To overcome this, the suggested system uses hybrid neural architecture that combines both convolutional and recurrent networks in learning both spectral and temporal features of spectrogram representations of audio signals. Varied set of samples of voice is gathered and pre-processed by means of normalization, noise removal and generation of spectrogram by STFT to guarantee good quality model inputs. Multitask learning is used to train the model so as to maximize both age and gender predictions. The use of accuracy, precision, recall, and F1-score to evaluate the methodology reveals a high level of performance, which proves the effectiveness of the deep learning methodology. The findings demonstrate the prospects of voice-based demographic forecasting in improving targeted advertising through providing more relevant and personalized user experience.