Posts Tagged ‘xab’

Video: German speech model ‘xab’

Friday, August 13th, 2010

I updated the video recognize-german.wmv (size: about 30 MB). In the video, you can see how I dictate more than 500 different German words with the German speech model “xab”. The space bars are now inserted automatically thanks to the dictation plug-in. The speech model has been trained with Ralf's German IPA FLAC files (section: xab). Each word has been trained exactly one time.

German speech model “xab”

Friday, August 6th, 2010

In this article, I will explain how I create the German speech model "xab". This speech model will contain about 758 German words.

import-pls-dictionary-xab1. I import the German PLS dictionary xab.xml.bz2. The result will become visible in the active vocabulary.

2. Grammar > Add sentence > Unknown
3. Commands > Manage plug-ins > Add > Dictation
4. Dictation > Append text after result > “[hit space bar]”
5. Training > Import training data > Import prompts. Paths on my computer:
/home/ubuntu/Documents/201006/audacity/xab-folder/prompts-xab (license: GPLv3)
/home/ubuntu/Documents/201006/audacity/xab-folder/wav
(license: GPLv3)

6. Export the current scenario. Then modify the scenario slightly with the text editor Geany. Then import the scenario.

7. Start ksimond. Simon > Press the Connect button. Then simon > Press the Synchronize button. Simon > Press the Activate button.

8. Dictate some words with simon:

abebbend abebbten Abendanfrage Abendanzug Abendblatt Abendessen Abendfriede Abendgebeten Abendkarte Abendmahl Abendmahlsgemeinschaft Abendprogramme Abendschauen Abendunterricht Abenteuerbuch abenteuerreichem Aberchen abundant abänderndem Abänderungsantrags Arbeitsstunde Arbeitsweise Arbeitszeitmodell Arbeitszimmer Aubings Rabenmutter raubendem Reibeisen Reibung Reiblaut Reibungsgewicht Reibungsverlust Schablone schablonierende Scheiben schraubenartige Schraubengangs Schraubklemmenleiste schreiben schreibfaule

Some recognition errors occurred, but most words were recognized correctly.

9. I want to offer the German speech model "xab". It seems that simon uses different paths. Here are the paths on my computer:

file:///home/ubuntu/.kde/share/apps/simon/model/hmmdefs
file:///home/ubuntu/.kde/share/apps/simon/model/tiedlist

A few seconds ago, I found out that the paths aren’t changed. I will use the following paths for the German base model "xab":

file:///tmp/kde-ubuntuJPP9Sy/simond/default/compile/hmm24/hmmdefs
file:///tmp/kde-ubuntuJPP9Sy/simond/default/compile/tiedlist
file:///tmp/kde-ubuntuJPP9Sy/simond/default/compile/hmm24/macros
file:///tmp/kde-ubuntuJPP9Sy/simond/default/compile/stats

Get these four files (hmmdefs tiedlist macros stats), and the corresponding scenario file (xab-scenario.xml): Download the German speech model "xab" (license: GPLv3).