CMU US KAL diphone database

This databases consists of a set of nonsense words containing all phone-phone transitions for US English. This database is free for any use (see licence for details). This database includes, waveforms, laryngograph (EGG) files, hand corrected labels, extracted pitchmarks, and various support files. It is released both as an example and in the hope you can make the festvox_kal voice sound better.

• Packed versions in bzip2 and zip format.
• Unpacked versions for your perusal, without requiring full down load. Note this is the actual waves, labs and pm used in building festvox_kallpc16k.tar.gz. The example directory is walkthrough of the diphone building code using autolabelling and hence doesn't have hand correction.
• Diphone building example This contains the results of a full walkthrough of the US diphone building example in the festvox document. You don't need to download this as you can construct it from the waveforms, scripts in the festvox distribution, and (optionally) the EGG signals

CMU TIME AWB limited domain database

This is a very simple example of a limited domain database for reading the time. This example is discussed in the "Limited Domain Synthesis" chapter of the festvox document. This database is free for any use (see licence for details). This database includes, 24 utterances autolabelled and built into a clunits synthesizer.

• Packed versions in bzip2 and zip formats.
• Full example This contains the results of a full walkthrough of the limited domain synthesis example in the festvox document. You don't need to download all of this as you can construct it from the waveforms, and scripts in the festvox distribution.
• A Run-time demo of the voice built from this databases is also available

CMU WEATHER AWB limited domain database

This database allows weather reprorts for the whole US. It uses the hourly information from http://weather.gov. The database consists of 100 reports automatically constructed to cover, date, time, outlook, temperature and wind direction. This database is free for any use (see licence for details). The weather reports and generation of the text of the reports is done by a simple script

• Full directory structure in
• A Run-time demo of the voice built from this databases is also available
• The clunits unit selection module for Festival used in these limited domain example sis still being updated. The actual version used in this example is availabel from festopt_clunits.tar.gz (though later versions also work).

CMU Communicator KAL limited domain database

This databases is used for a limit domain synthesis for the dialog system used in the CMU Darpa Communicator. Communcator is a automated telephone based dialog systems for booking flight information. This is much more general domain than time or weather information, but shows that an adequate voices may still be generated. As this voices is tailored to the particular language generation module in the CMU communicator The database consists of 500 utterances. selected by looking at the most frequent utterances made by the communciator over a 3 months period, plus others to ensure word coverage of the domain. This database is free for any use (see licence for details).

• Full directory structure in
• Packed version suitable for running and future development (wav/ lab/ pm/ festvov/ bin/ etc/ festival/ wav/ mcep/ src/).
• Am example dialog
• The clunits unit selection module for Festival used in these limited domain example is still being updated. The actual version used in this example is available from festopt_clunits.tar.gz (though later versions also work).