Instrument Synthesis

The Gameduino's sound system is a 64-voice synthesizer. Each voice generates a pure sine wave at a certain pitch and volume.

Using the sound processing package Loris, you can analyze a one-note instrument sample to generate a graph of how its key frequencies (or "partials") develop over time. This graph shows the 16 strongest frequencies in a single :download:`piano note <piano.wav>`, and how their volume peaks at the beginning of the note, then decays over the next second or so.

A little more code (Loris has a Python interface, fortunately) turns this note data into a form that the Gameduino can play directly through its 64-voice synthesizer. There are several of advantages of doing this instead of using a straight sample:

• control over pitch: to play at a different pitch, just scale the frequencies - duration is unaffected
• control over duration: by updating the amplitude values faster or slower, it is simple to lengthen or shorten a note
• reduced memory - the sample above would be about 10Kbytes as an 8-bit sample, as an instrument it takes about 600 bytes

The encoding is done by this Python program (analyze-instruments.zip) - it reads the note samples and writes Arduino data into a header file for each instrument. The format of this instrument data is:

• number of voices (1 byte)
• duration of note (1 byte)
• frequencies for all voices (voices * 2 bytes)
• amplitudes for all voices for all time steps (voices * duration bytes)

To play back the note, the Arduino sets up the frequencies for all used voices, then updates the voice amplitudes for each time step:

  PROGMEM prog_uchar *instr = piano;
  byte voices = pgm_read_byte(instr++);
  byte duration = pgm_read_byte(instr++);
  for (byte i = 0; i < voices; i++) {
    GD.voice(i, 0, (pgm_read_word(instr) * 184L) >> 10, 0, 0);
    instr += 2;
  }
  for (;;) {
    prog_uchar *amps = instr;
    for (byte i = 0; i < duration; i++) {
      for (byte j = 0; j < voices; j++) {
        byte v = pgm_read_word(amps++);
        GD.wr(VOICES + 4 * j + 2, v);
        GD.wr(VOICES + 4 * j + 3, v);
      }
      GD.waitvblank();
      GD.waitvblank();
    }
  }