A voice's gender is considered to be dictated by one's biology and cultural situation. Without modification, this determinism results in colinearity between acoustic metrics, making disentangling a metric's contribution to gender perception difficult. To study disentanglement on natural speech, we collaborate with a gender-affirming voice teacher to collect the Disentangled Source-Filter Dataset (DSFD): 45-minutes of audio along 25 Pitch, Resonance, and Weight voice configurations, coupled with Electroglottograph (EGG) measurements. Our analysis demonstrates certain acoustic and physical metrics, namely avg. F0, ∆F, Contact Quotient (CQ), and Loudness correlate with Pitch, Resonance, and Weight. Going on to perform perceptual studies of gender, naturalness, and realness, we see that ∆F is the strongest predictor of perceived gender. Perceived naturalness and realness of a voice, however, prove to be unpredictable by these acoustic metrics.