• You acquire theoretical skills to turn a linguistic research question (phonology) into the design of an experiment on a corpus of recorded speech (phonetics).
• You also acquire practical skills to perform such an experiment and report on it in an article.
• You will become familiar with some of the techniques used in speech signal analysis, synthesis and manipulation.

• Acquiring and managing a speech corpus.
• Formulating a linguistically inspired research question (phonology: segments, phonemes, discrete representations, the lexicon, arbitrary rules) that could be answered by a speech corpus (phonetics: durations, spectra, pitches, silences, noises).
• Designing an experiment on a speech corpus (selection of tokens and speakers, selection of analysis methods).
• Collecting the data.
• Techniques for speech analysis and synthesis.
• Basic statistical techniques for finding effect sizes (confidence intervals of differences and correlations).
• Automating the procedures for data collection and analysis in a Praat script.

• The book Experimental Phonetics by Katrina Hayward, Longman, London 2000 (available from the teachers during the first lecture).
• The book Speech signal processing with Praat by David Weenink.
• The latest version of the Praat program.

Techniques: corpora, software, research question, the mean
First we introduce a corpus and the analysis software. Then we introduce our first example research question (are there duration differences between the Dutch vowels /aː–ɪ–i/?). Then a statistical technique: computing the mean in four ways (by eye, by hand, with the Calculator, with a Praat script). Finally some programming techniques: expressions, variables, loops.
Homework 2: extract the tokens by hand for the eight speakers of the IFA corpus, and compute the three average durations of /aː–ɪ–i/, each in a different way.

Technique: standard deviation
First a discussion of homework 2.
Then another statistical technique: compute the standard deviation in three ways (by hand, with the Calculator, with a Praat script).
Homework 4: compute the three standard deviations of the durations of /aː–ɪ–i/, each in a different way.

Technique: sign test
First a discussion of homework 4. Then a statistical technique: factoring out much of the variation between speakers: a test nearly by eye (sign test on paired data); computation of its statistical significance with binomialQ (with the Calculator and with a Praat script).
Homework 6: compute whether /i/ is reliably longer than /ɪ/, and whether /aː/ is reliably longer than /i/, each in a different way.

Technique: the size of a difference
First a discussion of homework 6.
Then a statistical technique: to report how big a difference between two things is, we compute the frequentist confidence interval of this difference with invStudentQ (with the Calculator and with a Praat script); if this interval does not include zero, we have found a reliable difference (Student’s t-test on paired data).
Homework 8: compute how much longer /i/ could be than /ɪ/, and how much longer /aː/ could be than /i/, each in a different way.

Technique: reproducibility of your experiment
First a discussion of homework 8. Then a new method for automating the analysis of the experiment: annotate corpus data with TextGrids in Praat.
Homework 10: mark the beginnings and endings of the /aː–ɪ–i/ vowels in Praat (both replications), then write a Praat script to extract the results as a table.

Technique: tables
First a discussion of homework 10.
Homework 12: from the vowel duration experiment you have a headered table of data that can be queried for statistics by hand in SPSS, Excel, or Praat. Read it into Praat and retrieve the statistics (mean, standard deviation, sign test, t-test, confidence interval).

Technique: programming and t-test
First a discussion of homework 12.
Homework 14: how to write Praat scripts: automate the t-test.

Techniques: programming, and graphical display of data
First a discussion of homework 14.
Then a graphical technique: plotting vowel spaces in Praat.
Homework 16: measure the vowel space of one speaker of the IFA-corpus by hand, and use Praat’s history mechanism to draw it.

Technique: annotation
First a discussion of homework 16.
Homework 18: label all vowels of eight speakers of Dutch.

Technique: automated formant analysis
First a discussion of homework 18.
Then we introduce our second example linguistically-informed research question (are the auditory heights of front and back vowels correlated?)
Homework 20: programming, and plotting the vowel spaces of all eight speakers of the IFA-corpus by adapting your existing Praat script.

Technique: programming and pooling
First a discussion of homework 20.
Homework 22: correlation.

Technique: correlations
First a discussion of homework 22.
Homework 24: use and misuse of statistical methods.

Preparation for the final test
Homework 25: read Hayward's Appendix.