The Voice of an Instrument: Analysis of X-vectors for Music Emotion Recognition
Mariana Rodríguez Castañeda (UNAM), Iran R Roman (Stanford University)*
This paper will be presented virtually at the 12:15 PM - 12:45 PM PST session
Abstract:
This work explores the feature of x-vector to analyze the extent to which it embeds musical content by recreating the experiment in the work 'You’re Not You When You’re Angry: Robust Emotion Features Emerge by Recognizing Speakers' of emotion modulation in natural speech. As actors, we take instruments playing with four emotions; we use an autoencoder to observe changes in the reconstruction error, which allows us to find a similarity between the emotional modulation that occurred in speech in instruments.