Log on: Remember me
Powered by Elgg

timo :: Activity :: Just Me

People: Everyone | Friends & Community | Inbox | Just Me
Display: Full-text | Summary
Include: Blog Posts | Blog Comments | Files | Wiki Page | Wiki Comments

<< Back

Page 1 of 2

poetry generation using ASR and TTS
timo | weblog | Thu Nov 04

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 <sil> der hach<sil> also eben"
S (canned): "to my hooday, to my hooday"
U: "De masten so scheep as den Schipper sien Been"
S: "hm mach <sil> nach $t es in der <sil> 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
timo | weblog | Sat Apr 17

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 <http://www.gnu.org/licenses/>.

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
timo | weblog | Sun Oct 11
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
timo | weblog | Wed Jul 01
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
timo | weblog | Wed May 20
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
timo | weblog | Fri Jan 30

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
timo | weblog | Tue Jan 20

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
timo | weblog | Thu May 15

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...

eclipse plugins
timo | weblog | Wed Apr 16

I am currently investigating the ton of classes that implement the Sphinx interface "SearchSpace", or one of the three sub-interfaces. There are 19 in total and I am likely to have to add another one for the feature that I have in mind.

Anyway, I decided that I need something, preferably an Eclipse-plugin to visualize class dependencies, and there are actually a few options:

  • X-Ray would probably do the job, but it doesn't work. Maybe I just don't know how to install it correctly.
  • Byecycle have a great screencast on their page and more friendly installation instructions. It shows a dependency graph between classes and automatically and incrementally optimizes the graph layout. Infinitely. Using 20% of your processor(s). It's quite slow and it seems to be limited to only show dependencies within the package, while the dependencies I'm interested in often cross dependency boundaries (classes from different packages implementing an interface).
Also, I found Fat Jar, an Eclipse plugin that turns your whole project into a single jar. That's something my collegue asked my about the other day.
maɪ̯ iːpaː tɛkstfiːld
timo | weblog | Sat Apr 12

This has been programmed before, but here it is for you to see (and use):

IPATextField, a simple descendant of JTextField that will only accept phonetic input (either SAMPA or IPA if you know your uni codes by heart) and show IPA symbols.

You can try it out directly, as a main routine is included. It's even useful as your tiny copy-and-paste-IPA-editor.

<< Back

Page 1 of 2