Improving accuracy for classifying audio based mel frequency features using artificial neural network in comparison with Random Forest algorithm

The primary objective of the study is to explore the effectiveness of Audio classification using a Novel Artificial Neural Network (ANN) Algorithm in comparison with Random Forest to classifying different sounds into specific groups and representing them as waveforms. Material and Methods: The dataset which has been collected from Urban Sounds 8k that include a total of 8700 audios. Each Audio is composed of 40 rows and 40 columns for the classification of audio based mel frequency. Researchers compared the performance of a novel Artificial Neural Network (ANN) with a Random Forest algorithm for audio classification. A statistical analysis ensured a high probability of detecting true effects (power of 0.8) while minimizing false positives (alpha of 0.05). The ANN achieved a significantly higher mean accuracy (94.34%) compared to Random Forest (67.26%) over 10 trials (p-value = 0.004). This suggests that ANNs are a promising approach for audio classification tasks based on the findings of this study.