Skip to content
SpectrogramPrivate
enid

Build a sound

Stack pure tones and listen to what you have made. Then check the picture underneath: the transform should find exactly the tones you put in, at exactly the strengths you chose. You supply the right answer, so you can test the tool rather than trust it.

Your tones

3
  1. Tone 1
    220.0Hz
    0.60
    0.00π
  2. Tone 2
    440.0Hz
    0.30
    0.00π
  3. Tone 3
    660.0Hz
    0.15
    0.33π

The wave you built

What the transform found

20 Hz50 Hz100 Hz200 Hz500 Hz1 kHz2 kHz5 kHz10 kHz20 kHz

Does it come back?

Run the transform, then run it backwards. If it is correct, the signal that comes out is the one that went in — to the last decimal a computer can hold.

Worst deviation—
Energy kept (Parseval)0
#PitchYou setFound
1220.0Hz0.6000.000
2440.0Hz0.3000.000
3660.0Hz0.1500.000

The strengths found come out a shade under the ones you set. That is not an error: a tone whose pitch falls between two bins is shared across both of them, so neither reads its full height. Land a tone exactly on a bin — 187.5 Hz here — and it comes back whole. Peak: 1.004

Fine-tune
Window function

Blackman — least leakage, widest peak

Cutting a signal off abruptly invents frequencies that were never in it — leakage. A taper is how that is kept down.

Frequency scale

Linear is what the FFT returns, log matches how pitch is heard, mel matches how hearing spaces it out.