Pitch Detection Using a Tunable IIR Filter

instruments. Except for the electronic keyboard, where each note is detected via a switch closure, few musical instruments are readily adapted to even the simplest of sound synthesis algorithms. The problem lies in the difficulty of detecting and tracking the pitch (note) sequence in real-time as the instrument is played. Standard spectrum analysis techniques, although very useful in general, are not well suited to solving this problem, as one may find when attempting to apply the Fast Fourier Transform (FFT) algorithm to the pitch-following problem. For a tutorial on the FFT, see Jaffe (1987). The FFT divides the frequency domain into equal intervals on a linear scale such that the frequency difference Af (or resolution) is the sample frequency fs divided by the number of points N comprising the sample time window. The FFT sample time window (where the pitch should ideally remain constant) is the inverse of the frequency difference Af. In order to obtain a resolution of one cent, where a cent is defined as 21/1200 such that there are 1200 cents per octave (Lancaster 1974), the number of points required in the FFT is then at least N = 4800 to allow 1200 points to span one octave from f,/4 to fs/2. The total time needed for a stable signal (the FFT sample time window) is then 4800/f,. For a pitch ranging from middle A (440 Hz) down one octave to 220 Hz, f, = 880 Hz and the total time window is 4800/880 = 5.45 sec. Musical pitch data changes much more quickly than that-approximately on the order of 100 msec or faster for a rapid note sequence. Some of these problems may be overcome by utilizing a moving FFT where the sample windows slide through time creating multiple overlapping windows (Tuteur 1988), but this is done at the exPitch Detection Using a