pygame.mixer.Sound.play is irregular although fire

Question

I currently try to repeat a sound every x ms - where x is dependent on an UDP packet I receive via socket - and I decided to use pygame for that. I used this SO answer to repeat something every x ms: https://stackoverflow.com/a/18954902/3475778

But now I have the problem, that the sound is played very irregular and made a minimal-working example where the problem persists:

import pygame
from pygame.locals import *

pygame.mixer.init()
sound = pygame.mixer.Sound('sound.wav')

def play_sound():
    sound.stop()
    sound.play()

pygame.init()
clock = pygame.time.Clock()

pygame.time.set_timer(USEREVENT+1, 200)

while True:
    # clock.tick(30)
    for event in pygame.event.get():
        if event.type == USEREVENT+1:
            play_sound()

Here is the waveform of what I have recorded from the script via Audacity:

You see that for some reason some samples were played longer than the others. Not very nice for some kind of metronome I want to build.

edit UPDATE: It is not a problem of pygame.time.set_timer, because this code doesn't solve the problem and doesn't rely on pygame.time.set_timer:

import pygame
from datetime import datetime

d = datetime.now()

pygame.mixer.init()
sound = pygame.mixer.Sound('horn_short.wav')

pygame.init()

while True:
    if (datetime.now() - d).total_seconds() > 0.2:
        sound.play()
        d = datetime.now()

has the same problem. The Problem is also under Ubuntu 16.04, Python 3.5 64bit (Anaconda) and a fresh installed pygame.

Answer 1

For me the thing worked much better, if i did:

pygame.mixer.pre_init(44100, -16, 2, 256)

before any of the pygame init functions.

Answer 2

Here is an idea for an alternative approach. If the goal is to play a sound in regular intervals, you might get better results if you (dynamically) pad the sound to the desired interval length, and then simply loop it with Sound.play(loops=-1).

If there are just a handful of valid intervals, it might be easiest to store dedicated sound files and load the appropriate one.

Otherwise, the pygame.sndarray module provides access to a numpy array of the raw sound data, which makes it possible to dynamically generate sound objects of the desired length:

import pygame
import numpy


# Helper function
def getResizedSound(sound, seconds):

    frequency, bits, channels = pygame.mixer.get_init()

    # Determine silence value
    silence = 0 if bits < 0 else (2**bits / 2) - 1

    # Get raw sample array of original sound
    oldArray = pygame.sndarray.array(sound)

    # Create silent sample array with desired length
    newSampleCount = int(seconds * frequency)
    newShape = (newSampleCount,) + oldArray.shape[1:]
    newArray = numpy.full(newShape, silence, dtype=oldArray.dtype)

    # Copy original sound to the beginning of the
    # silent array, clipping the sound if it is longer
    newArray[:oldArray.shape[0]] = oldArray[:newArray.shape[0]]

    return pygame.mixer.Sound(newArray)


pygame.mixer.init()
pygame.init()

sound = pygame.mixer.Sound('sound.wav')

resizedSound = getResizedSound(sound, 0.2)
resizedSound.play(loops=-1)

while True:
    pass

Using just pygame (and numpy), this approach should have a good chance for an accurate playback.