It's been claimed that they can read, as they have laughed at posts in this thread. However, when it came to testing their ability to read, it turns out that they can't.
Perhaps there's some way we could test their ability to change sound files after they've been recorded? This could be done by using an external recorder to record the sounds of playback, then play the same clip back several times, with flaccon confirming that the answer was different each time.
This would have to be done somewhere where total silence is possible and where the recording could be very controlled, so there would need to be the hire of a recording studio, or something similar. Then the waveforms of the recordings could be compared visually to see how different they are, if at all.
Seems rather expensive, and reliant on what will end up being subjective criteria, but if the sounds are easily discernible and truly contain different words, then the waveforms should look markedly different. You could even compare what the spirits are saying with a waveform of someone else saying the same thing. I've done a fair bit of speech and vocal editing in my time, and different words have recognisable shapes, even when spoken by different people with different accents. So there should be marked similarities between the waveforms generated by recording the spirits and a waveform generated by recording someone saying the same phrase into a microphone.