Speech analysis with Praat
Paul Trilsbeek
DoBeS training courseJune 2007
Sources used
• Sidney Wood’s “Praat beginners’ manual”
• Pascal van Lieshout’s “Praat short tutorial”
• Links for both on www.praat.org
Outline
• First half: theoretical overview– What is Praat– General Praat interface– Several kinds of analyses– Creating graphical output
• Second half: practice– Hands-on experience with some exercises
What is Praat? (1)
• “Praat” = “speak” (Dutch)• “Doing phonetics by computer”• Analysis and reconstruction of speech
signals• Developed by 2 phoneticians from the
University of Amsterdam, Paul Boersma and David Weenink
What is Praat? (2)
• Program running on most operating systems (Windows 95 to XP, Mac OS 9/X, Linux, several kinds of Unix …)
• Available for free from www.praat.org under GNU General Public License (inc. source code)
General Praat interface
• Praat objects window– List of praat ‘objects’: sounds, analyzed
sounds, synthesized sounds …– Buttons for editing, viewing, applying
analyses, etc.
• Praat picture window– For creating and manipulating graphical
output
General Praat interface
Getting help
• Praat help menu• Praat web site (same content)• 2 tutorials mentioned before (see
praat.org)• Praat discussion list
Recording a sound
Loading a sound
• Long vs. short sound file: memory• Extract part• Extract segments from annotated long
sound
Annotating a sound
• Create text annotations on different levels
• Praat objects can be created from segments
Annotating a sound
Praat Edit window
• Waveform display• Segment play buttons• Zoom buttons• Several other views of analyses
– Spectrogram– Pitch analysis– Intensity analysis– Formant analysis– Pulses
Praat Edit window
Spectral analysis
• Fourier theorem: “any given signal can be constructed by addition of an infinite number of sine waves with specific frequencies, amplitudes and phases”.
• Fourier analysis: “ finding out the frequencies, amplitudes and phases of these sine waves for a given signal”
Spectral analysis
Spectral analysis
Spectrogram
• Spectrum with a third dimension added: time.
Pitch analysis
• Shows fundamental frequency where one can be detected
• Not always correct, octave jumps often occur, depending on the quality of the signal. Can be corrected manually.
Pitch analysis
Intensity analysis
• Shows the intensity contour of a signal
Intensity analysis
Formant analysis
• Frequency bands with higher acoustic energy caused by resonances in the vocal tract, mouth and nose cavities
• Different vowels and fricative consonants have different formant structures
Formant analysis
Manipulation
• Add or delete pulses• Modify duration (timing)• Modify pitch contour
Manipulation
Picture window
• Clicking “draw” button will draw graphic into picture window
• Some modifications can be done in the picture window
• Text can be added
Picture window
Picture window
• Export picture as EPS (Encapsulated PostScript) for further use in media to be printed. Adobe Illustrator can be used to further edit the EPS file
• Export picture as Windows Meta File (Windows only) for media to be shown on screen (PowerPoint presentation)
Picture window
• EPS output:
Picture window
• WMF output:
Exercises