Hmmm. I know for a fact that Audio::Play (should you decide to use it) uses floats for sample data and most pcm output hardware uses 16-bit words per sample. Your code assumes its 8-bits.

It's also probably more efficient to generate an @array of data first (say 100 samples) and then pack() it into a string and write it (Audio::Play can do this for you):

#!/usr/bin/perl -w
use strict;
use Audio::Play;
use Audio::Data;

my $player = Audio::Play->new();
while (1) {
    my $audio = Audio::Data->new( rate => 44100 );
    my @samples = map { sin ( 2 * 3.14 * $_ / 100 ) / 8 } 0 .. 99;  # 
+generate 100 samples as perl numbers
    $audio->data(@samples); # will pack() samples into floats
    $player->play($audio);
}
[download]

This works perfectly on my machine.

Update:

To clarify: you really don't want to calculate and then play each sample seperately. It's incredibly inefficient so even most C - based softsynths don't do that.

What you want is a low latency between dragging a slider, and hearing the sound change. IMHO checking the slider state once for every 100 samples is enough:

Say you're running at 44100 Hz (normal CD sample rate), then playing 100 samples takes 100/44100 =~ 0.0023 seconds. That's probably fast enough for anything except really exact keyboard input, provided you don't have to sync your ouput with audio streams generated from outside your machine.

So, what you'd do is (pseudocode):

while (1) {
  check_for_input();
  update_parameters_for_new_input(); #possibly using heavy math
  generate_100_samples_as_fast_as_possible();
  push_samples_into_soundcard_buffer();
}
[download]

Hope this clarifies.
Joost.