I will likely have some post coming up that were delayed due to my long period of silence. This post falls under this category.
Some time ago, one student pointed me to this video of a talking piano. The video is in German but the piano talks in English (the subtitles in the video help to understand what the piano says).
The details of the video explain that the procedure to get the piano to talk started with a recoding of a child. The creator of the talking piano calculated some kind of spectrogram of the speech and then translated the time-frequency information in the spectrogram into sequences of notes played in the piano. One interesting thing to note here is that typical spectrograms, based on the Fourier Transform, will give linearly separated frequencies, but the frequencies generated from the piano keys are logarithmically separated. Maybe using wavelets would have been better, but there is no information if this was the case. Another interesting thing to note is that the piano implements parts of the processing done at a vocoder or also at a sub-band coder.
