How do I get rid of this crackle/pop noise when my waveform program goes silent?

Crackles and pops at the start or end of a waveform are usually caused by sudden jumps in the audio signal. In digital audio, every abrupt change introduces high-frequency components that your ears hear as clicks. This is especially noticeable with 8-bit audio, because the signal has fewer levels, so even small transitions become abrupt.

The common ways to fix this are:

1. Fade-in at the start and fade-out at the end - instead of starting or stopping the waveform instantly, gradually ramp the volume up and down over a few milliseconds.

2. Align waveform stops to zero-crossings - stopping exactly when the signal crosses zero reduces pops, but fade-in/out is usually simpler.

3. Use higher bit-depth - 16-bit audio gives smoother steps and reduces audible clicks.

Here’s an example for fade-in/fade-out in your current code:

def applyfade(samples, fadesamples=50):
    # Fade-in
    for i in range(min(fadesamples, len(samples))):
        samples[i] = int(samples[i] * (i / fadesamples))
    # Fade-out
    for i in range(min(fadesamples, len(samples))):
        samples[-i-1] = int(samples[-i-1] * (i / fadesamples))
    return samples

You can call this on your data array before writing it to PyAudio:

data = bytearray(applyfade(data, fadesamples=100))  # Adjust fade length as needed
channel.write(bytes(data))

A few practical tips from working with PyAudio:

• Keep the fade short but not too short; around 50–200 samples is usually enough for 16 kHz audio.

• Always apply fade to every waveform chunk that might start or stop suddenly.

• Using paInt16 instead of paUInt8 makes fades smoother and reduces stepping artifacts.

These changes remove most pops and clicks without affecting the waveform noticeably. It’s not a PyAudio bug; it’s just the physics of digital audio.

I was able to figure out what went wrong, and thought I'd post here in case others have the same issue.

The actual problem was endemic to my method of generating the tone--since I was using unsigned 8-bit integers, the middle point was not 0, but rather 128. I had not compensated for this in my code, and that was causing the pop to be inevitable. Fixing the problem was as simple as using the generated wave value as an offset from 128 minus half the volume (i.e. if the volume is 4 and the pulse wave generates 0 or 4, the output should not be 0 and 4, but 126 (128 - volume/2 + 0) and 130 (128 - volume/2 + 4)).