Top panels: Two tonal sounds with fo of 150 Hz (left) and 550 Hz (right) are filtered by B middle panels: the same transfer function corresponding to vowel [a] spoken by a person with a 15.7 cm long vocal tract (c = 354 m/s). C and D (left panels): dense harmonics of the 150 Hz vowel clearly reveal spectral peaksputative formants F1 to F4 that closely correspond to the true vocal tract resonance frequencies. In contrast (C and D; right panels), in the vocal signal with fo of 550 Hz, we also observe spectral peaks (P1, P2, ); but these now correspond to harmonics of fo, deviate considerably from the true vocal tract resonances, and should not be confused with formants. Note that although the fundamental frequency is typically partial with the highest amplitude in the initial source signal, higher harmonics may be stronger in the output signal after filtering. Diagnostics: harmonics are always spaced at exact integer multiples of fo, whereas formants can vary independently of one another and thus are rarely perfectly evenly spaced
