spectral-features

Audio feature extraction and classification

In this work is proposed a speech emotion recognition model based on the extraction of four different features got from RAVDESS sound files and stacking the resulting matrices in a one-dimensional array by taking the mean values along the time axis. Then this array is fed into a 1-D CNN model as input.

The code shows implementations of basic features for voice/speech analysis.

A compact 8‑channel EEG method for depression screening is evaluated strictly at the subject level, revealing that true accuracy is much lower than segment‑level reports.

Cross-corpus evaluation of spectral–acoustic feature fusion for pathological voice discrimination using the MEEI and USP voice pathology corpora.

Corrected evaluation of temporal energy descriptors for speech arousal estimation on RAVDESS. Supersedes an earlier version whose results came from a leaky split — see CORRECTIONS.md.

This repository presents the results of a technological initiation that encouraged my undergraduate thesis, I sought to improve my knowledge in digital signal processing applied to music. It is a spectral analysis tool for the audio of musical instruments, focusing on objectively characterizing timbre.

Multi-input DenseNet architecture with data augmentation (sliding window/overlap). Published in IEIE (2022).

Streamlit app for frequency-domain audio signal analysis — FFT, spectrogram, spectral features (centroid, bandwidth, SFM, SCF), formant detection, and F0 estimation via cepstrum.