Ok, since pcmplay has been integrated into ttsd, testing was a little more convoluted....but;

I installed ttsclock to get some nice pcm samples. I then installed pcmplay.overlay, compiled from your source.

Testing with cat /drive0/var/tts/evening.pcm | /programs0/pcmplay.overlay gave the same results - volume spikes.

I then commented out the call to audio_init() in pcmplay.overlay.c, recompiled and reinstalled. No spikes - this is smooth, and the way to go.

One problem with this is that it only works when pcm is already being output. ie, if the player is paused, or in AM/FM/Aux mode, then nothing will be output. (I'm guessing that Soft Audio Mute is set in paused mode) This is an issue, but the original pcmplay code didn't do the right thing anyway - What really needs to be done is;

Get the current source and store it.
Get current SAM status and store it.
Set SAM and change to PCM (if neccessary.)
Unset SAM and pcmplay the sample.
Set SAM and change back to stored source. (if it was changed)
Restore SAM to original state.

This should allow pcmplay to always output files, regardless of current source and mute settings (and restore them afterwards). One could further question whether we'd want to look for a *minimum* volume level that should be active for samples...eg if we've turned the volume right down to have a conversation, and we want GPSapp to announce a turn, should we turn the volume up a bit?
This should be done in a similar way - check and store the current volume, and if it's below a certain level (preferably configurable...but we're only feasibility testing at the moment...can be done later), raise the volume to that level, and flag that we changed it. Do everything else, and then if the flag was set, restore the previous value.

As long as any volume changes are done via the existing ioctl's in empeg_mixer.c, volboost shouldn't conflict in any way. The only way that volboost can really get screwed up is if something within the kernel tries to set the volume directly without taking volboost into account. I don't see any need for this to be the case for pcmplay.

@TheAmigo; Can you comment out the audio_init() call in ttsd?