Now look at the third waveform. This is the same audio as in the first two waveforms but with the polarity flipped. Notice that now with 100 percent modulation on the negative peaks, the positive peaks do not even reach 100 percent (about 80 percent). This is the undesired setting for the audio polarity.
Except in the case where the modulation capability of the transmitter doesn't exceed 80%. This could be due to deficiencies in a plate modulated transmitter, or a linear/grid modulated amplifier running at too high carrier efficiency.
Back in the 1930's there was some recognition of asymmetrical voice waveforms, and one school of thought was to set the carrier efficiency at something like 50%, and run the audio polarity so that the greater amplitude peaks extended in the negative direction. It was argued that with the same tubes and power supply, more sideband power would be generated than when running at lower carrier efficiency but letting the greater amplitude peaks extend in the positive direction.