Voice Emotion

This data source uses a Speech Emotion Analyzer demo based on the formantanalyzer JavaScript library. This algorithm extracts syllables from the Mel-spectrogram of the input and feeds relevant statistical features into a single-layer neural network that classifies. This model was trained on datasets of actors going through dialogues or improvising. It predicts the following categories:

• Happiness

• Sadness

• Anger

• Neutral

From these labels, we output valence and arousal scores by using the relative happiness probability (for valence) and neutral probability (for arousal).

Attribution

GitHub - tabahi/WebSpeechAnalyzer: JS speech analyzer for fast speech analysis and labelingGitHub