• Rahul Gupta
• Saijal Agarwal
• Ravina Rewar

Deep learning models are good at seeing but not so good at hearing. So, when it comes to audio or speech data, we don’t get expected results. Basically, our need was to train a deep learning model, which is able to classify the speeches on the basis of various characteristics of the speaker. In this paper, we tried to visualize differences among speakers on the basis of his Liveliness, Speech Rate and Vocal Depth in the form of images. We did this by taking the audio signals to spatial domain by using a method called Short Term Fourier Transformation. To better visualize the differences we used concept of Mel Spectrograms, which better correlates with human hearing. And we successfully visualized the differences with the help of mel spectrograms. And these differences were very significant and considerable which can be observed by anyone. These observations can be proven very useful in training deep learning models that can evaluate speakers.

Spectrograms, Mel Spectrograms, CNNs, Characterizing Speakers

"Characterizing Speakers Using Spectrograms", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.6, Issue 5, page no.164-168, May-2019, Available :http://www.jetir.org/papers/JETIRBW06034.pdf

"Characterizing Speakers Using Spectrograms", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.6, Issue 5, page no. pp164-168, May-2019, Available at : http://www.jetir.org/papers/JETIRBW06034.pdf