Hi folks: Brandon McGinty and I have been poking around attempting to fix the gravelly voice on Raspberry PI machines. The effect shows up on both the 32-bit and 64-bit systems. It took a bit of experimenting to find what I believe is the correct fix. The problem appears to be the standard buffer size being handed off to the audio subsystem finishes playing more quickly on the PIs causing the speech to not sound smooth giving a gravelly speech output. In the new espeak-ng and libespeak-ng this is handled by pcaudiolib instead of port-audio under the older espeak package. To fix the problem you have to hand over larger buffer of sound to smooth out the effect. The lines in question are line 103 in pcaudiolib/src/alsa.c and line 83 in pcaudiolib/src/pulseaudio.c. The default lines each end with dividing by 1000. I first tried just removing the / 1000 which smoothed out the speech very nicely but caused the buffers to be large enough that moving between lines and words in review mode to have an extremely sluggish response because the buffer wouldn't flush until it finished speaking it. So I put the / 1000 back in and started reducing the value to find the result that would provide smooth speech but still maintain responsiveness with the review functions of speakup. What I found worked fairly well was a division value of 200 which seemed to increase the buffer sizes enough to give smooth speech and keep the quick feel of moving around. If you are experiencing gravelly speech I'd appreciate it if you would try the above changes to see if it resolves the problem. You want to edit 1000 to 200 on lines 102 of alsa.c and line 83 of pulseaudio.c in pcaudiolib/src and then recompile. Different distros seem to put their libraries in various weird and wonderful places so be a tad careful wheninstalling the new libpcaudio libraries and symlinks. Even the RPI 32-bit and RPI 64-bit stash their libraries in different places. I hope this helps some folks. Kirk
