timo : RSS Feed

Comment on "aggregated audio duration"

01/26/12 05:19

Why did I only see this now?! This is tons useful, thank you. I wonder if I can cite this as a reference for a custom essay that I'm writing? I've been looking for a code that I can use for my essay and this one fits the bill. It's okay if you won't allow it, though, I'm just pushing my luck here.

Again, thanks for sharing this script!

Comment on "howto: SRILM and ubuntu"

02/12/11 20:51

http://www.google.com

poetry generation using ASR and TTS

11/04/10 09:38

how about the following: you feed the first verse of your favourite poem/song lyrics into an ASR and have a TTS read back what the ASR understood. It is very likely to rhyme! If you also read back a nice chorus, people will love it. Now, you just need your TTS to do decent singing and you're set:

U: "Ick heff mol in Hamburg een Veermaster sehn"
S: "gegen $l aber hin der hach also eben"
S (canned): "to my hooday, to my hooday"
U: "De masten so scheep as den Schipper sien Been"
S: "hm mach nach $t es in der warm $w"
S (canned): "to my hooday, hooday ho."

By the way, this is how singing works in a soccer stadium and the principle behind Chinese whispers.

aggregated audio duration

04/17/10 12:55

I've finally written the one script that was missing from the interwebs and that I have longed to have for so long:

#!/usr/bin/perl
# Copyright (C) 2010 Timo Baumann
# This program is free software; you can redistribute it and/or modify it 
# under the terms of the GNU General Public License as published by the 
# Free Software Foundation; either version 2 of the License, 
# or (at your option) any later version.
# 
# This program is distributed in the hope that it will be useful, 
# but WITHOUT ANY WARRANTY; without even the implied warranty of 
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. 
# See the GNU General Public License for more details.
# 
# You should have received a copy of the GNU General Public License 
# along with this program; if not, see .
use strict;
use warnings;
use Audio::Wav;
use Audio::Wav::Read;

#usage: audio-duration.pl path-or-file1 path-or-file2 ...

my @files;
for my $arg (@ARGV) {
    my $findresult = `find $arg`;
    push @files, grep /.wav$/, split " ", $findresult;
}
#print join " ", @files;
my $duration = 0.0;
my $wav = new Audio::Wav;
for my $file (@files) {
    my $read = $wav->read($file);
    $duration += $read->length_seconds();
}
# convert to something readable
my $readableDuration = "";
if ($duration > 600) {
    my $seconds = int($duration + .5);
    my $minutes = int($duration / 60);
    $seconds -= $minutes * 60;
    my $hours = int($minutes / 60);
    $minutes -= $hours * 60;
    $readableDuration = "(" . ($hours > 0 ? "$hours:" : "") . "$minutes'$seconds\") ";
}
print "$duration seconds ", $readableDuration, "in ", ($#files + 1), " wave files.\n";

Running this in any directory wil yield the duration of audio (only .wav files) of all the files in this directory. If you supply arguments, it will look into the given directories (or files) and tell you the summed duration.

A must-have for any corpus-linguist dealing with loads of audio files!

Update

10/11/09 19:40

I've been back from Stockholm for a while now, gone to Interspeech, SIGdial and YRRSDS and am now back at work in Golm. I'm now again (professionally) centered around two things: Getting our next prototype of an incremental SDS up and running in the next few months and continue to work on my PhD thesis, which I hope to finish some time next year.

Great new Sphinx Extensions

07/01/09 09:09

I finally got around to package the pitch tracker and some of our incremental result filtering (which was the reason for me travelling to Boulder, USA) as add-ons for Sphinx. Find them on my website.

Here and there

05/20/09 08:43

I have safely arrived in Stockholm, where I will be visiting researcher at KTH's speech lab for the summer. Also, I will be travelling to NAACL-HLT 2009 in Boulder to present our paper and a smaller workshop in Bielefeld. Also, I will participate in Dialholmia as a student volunteer. Many chances to meet and greet!

Speed-Up Boot under Ubuntu with VMware

01/30/09 09:50

I still need Windows for one piece of software that I use occasionally. So, since I've moved to Ubuntu, I've been using VMware for this (as it was the only solution at that time).

My VMware stopped working under Hardy. Luckily, I never needed my Windows-App for half a year. In Intrepid, I was able to just install vmware from their webpage and it restored my windows session from a year ago. Probably a record-breaking uptime for windows...

So, yesterday I played around with bootchart and found out, that the VMware-services took 4 precious seconds of my (and my battery's) lifetime on every boot. Not really worth it, as I'm unlikely to use my Windows-App anytime soon. So, here's what I did:

remove the links in /etc/rc*.d/*vmware

as the first command in /usr/bin/vmplayer add:

gksu -D "Need root priviledges to start vmware services." /etc/init.d/vmware restart

Works like a charm.

Generic Minimum Edit Distance of lists in Perl

01/20/09 15:26

The title says it all: I am looking for a generic implementation that tells me the edit distance of two lists. The implementations on CPAN all seem to work on string-data. Which is OK for finding typos but makes WER calculation tedious.

So, I want a generic implementation that takes a comparator-function (as in sort {$a <=> $b} @list) and two lists and outputs the edit distance. Nice to have would be distance-weights and really nifty if the value of the comparator function (not only !=0 but how much lower or higher) was taken into account.

Luckily I don't need it now, so I don't have to write it. But it would be a great finger exercise for a Perl-in-NLP class.

steps towards German ASR

05/15/08 17:48

Klotz hin Gnubbel

This is what I get with the current acoustic model and a LM that was even trained including the correct sentence (und füge es ein in den Bauch des Elephanten).

Even using only just the correct sentence as a grammar returns und füge es, instead of the complete sentence. The alignment shows, that es is supposedly spans the complete ein in den Bauch des.

I read that the current models are severely overtrained on one speaker, so I tried one of his utterances (de43-01, die Anwendung wird entwickelt) which is correctly understood if I use it as a grammar (effectively resulting in forced alignment) and which results in the beautiful phantasie wird entwickelt if I include this one sentence in the statistical LM as above.

Thus, the bad results are probably due to the bad acoustic model. I've already uploaded the PentoNamingCorpus to Voxforge, thus hopefully, acoustic models will improve eventually. But if bad comes to worse, we'll have to train based on KCoRS and Verbmobil...