From testing in SameBoy (I'm assuming it's accurate), this appears to depend on model-specific APU behavior. The speech plays when emulating CGB-C or earlier but not when emulating CGB-D or CGB-E.
Looking at what the game is doing, it's playing PCM samples through the pulse channels by very rapidly adjusting their volumes. It's re-triggering rapidly enough that the channels will not advance through their pulse waveforms while it's doing this, so the sound will only play if the waveform is stopped on a 1 rather than a 0.
The speech plays correctly if I mess with the pulse wave phases so that they end up stopped on 1s at the title screen, but this will require some further investigation to figure out the correct behavior.
From testing in SameBoy (I'm assuming it's accurate), this appears to depend on model-specific APU behavior. The speech plays when emulating CGB-C or earlier but not when emulating CGB-D or CGB-E.
Looking at what the game is doing, it's playing PCM samples through the pulse channels by very rapidly adjusting their volumes. It's re-triggering rapidly enough that the channels will not advance through their pulse waveforms while it's doing this, so the sound will only play if the waveform is stopped on a 1 rather than a 0.
The speech plays correctly if I mess with the pulse wave phases so that they end up stopped on 1s at the title screen, but this will require some further investigation to figure out the correct behavior.