As an example, I've written a bit of code here. The result is almost always longer in actual characters, but considerably shorter in syllables. I've restricted the consonants to try to stay with pronouncable monosyllables. Basically, each word is a combination of a consonant cluster from $consonants1, a vowel from $vowels, and a final cluster from $consonants2. You can add anything you want to those, as long as they stay aurally distinct. Or if you find some of them are hard to pronounce or keep distinct, you could remove any that you have trouble saying. As given, it'll convert "4345317546" to "stut-prip-bluv" and back.

#!/usr/bin/perl -l
use strict;
use warnings;
 
my @consonants1 = qw/ b bl br ch d dr f fr g gr h j k kl kr l m n
                      p pl pr r s sk skl skr sl sn sp spl spr st
                      str t tr v vr z /;
my @vowels = qw/ a e i o u /;
my @consonants2 = qw/ b bs d dge ds f g gs ck ll lf m n nd nt p r
                      s st t v x z /;
 
my $nwords = 0;
my @num2word;
my %word2num;
for my $c1 (@consonants1){
    for my $v (@vowels){
        for my $c2 (@consonants2){
            $num2word[$nwords] = "$c1$v$c2";
            $word2num{"$c1$v$c2"} = $nwords;
            $nwords++;
        }
    }
}
 
while(<>){
    chomp;
    if (/^\d+$/){
        my @words;
        #encode
        while ($_ != 0){
            push @words, $num2word[$_ % $nwords];
            $_ = int ($_ / $nwords);
        }
        print join "-", @words;
    } else {
        #decode
        my $num = 0;
        for my $word (reverse split /-/){
            $num *= $nwords;
            $num += $word2num{$word};
        }
        print $num;
    }
}
[download]