There are two ways that the sound may become muted. Firstly, the
power supplies to the VS1003 don't include a negative voltage
source. That usually means that the 'ground' is not in the
midrange of the AC output signals, and only a coupling capacitor
(to block DC) will make a suitable connection to an external
device (amplifier input or speaker). The DC bias can cause
an amplifier to saturate (hit its voltage rail, outputting only
a constant voltage with negligible signal content).
That DC-bias problem is not consistent with the observation that
the sound can be heard after removing one channel (and if the
amplifier/speaker on the output has GBUF biased correctly,
there is no need to modify the circuit as shown). GBUF should
correct this problem, but that depends on the common-mode
range of the output amplifier, so a DC measure of both the
absolute voltage, and GBUF voltage, is advised. The
other way that combining stereo signals can result in zeroing
the output, is if the 'stereo' is in fact monaural, and
the two channels are in opposite phases (basically, different
polarity copies of the SAME signal).
Check the outputs for DC content, and if you have an oscilloscope,
put left channel on X axis, and right channel on Y axis, and see if
the display is a diagonal line, rather than the more nuanced squiggles
of a known stereo sound source.