Homework #4

S702 D. Kewley-Port, W. Su2/4/02

This is the second homework on speech synthesis. The purpose is to become familiar with the KLSYN synthesizer. This contains the parameters for the KLTSYN88 synthesizer which has enhanced control over the glottal source compared to the KLATT80 synthesizer. KLSYN was written by Prof. Keith Johnson and Prof. Ying-Yong Qi.

1) Make sure all klsyn files are copied from the zip file from Oncourse into a directory in your cfs account. Please make a new folder for klsyn files. Use WordPad to read Readme.txt and default.con. For ee.doc, this is an ascii file that you can directly open the file by double click the file name. After the window comes up, choose "plain text" and the default and the parameter for ee will come up. The manual is in both klsynman.doc and klsynman.html, or on the web from our Acoustics Software link (click here).

2) Launch "command prompt" window under Programs->Accessories on Windows-NT. Find the mapped letter for your CFS (M: for me). Type that letter (M:) to change to the cfs directory. Then cd klsyn (or your folder name) to your  folder.

3) Start the synthesizer by typing: klsyn ee [Note: do NOT type .doc]
When you start KLSYN, it will look for 'default.con' in your CFS directory. Start KLSYN with an existing file like ee.doc to get the correct parameters for .wav files: Type: klsyn filename in the window

4) KLSYN documentation suggests that you should have a spectrogram making program and a listening program to help with the synthesis. You should use CoolEdit for this since KLSYN generates .wav files, but has no option for listening to them. You can use WordPad to view the parameter files such as ee.doc. However, if you want to synthesize new .wav and .doc files in KLSYN, you must either save them with a new name, or close your CoolEdit and WordPad files before you synthesize or the command prompt window will crash.

5) Work with someone from Lab #1 who made spectrograms. Select either /hed/ or /hehd/ and pick one of utterance from each talker to synthesize using klsyn. It is recommended that you start with one talker, and achieve a reasonable match to the original speech given the vowel measurements you made in Lab #1. To help with synthesis parameters for /d/, see p. 986-987 of the article and as well as the spectrogram of your utterance. To synthesize /h/, use the vowel formants but excited with AH instead of AV. It may be useful to check out the Klatt parameter book from Kim in the C100. Then modify the vowel in your synthesis parameters for taker #1, according to measures from Lab #1, so that it sounds reasonably like talker #2.

6)When you are reasonably satisfied with your synthesis, print out your spectrograms of the original and synthesized utterances. Also print out the .doc parameter file from the synthesizer for each talker. Comment in just a few sentences how successful you were in matching the talkers vowels. Post the .wav file for your own synthesized speech with your initials on Oncourse>In Touch>Group Files>klsyn. The best synthesized utterance (according to DKP and WSU) will win a small prize.