Rob Hagiwara's Monthly Mystery Spectrogram Webzone

Welcome to the Monthly Mystery Spectrogram webzone. These pages are Rob Hagiwara's professional web-space. For personal musings, please see Rob's blog.

to Current Mystery to Current Solution to Archives to How To Page to Research Page to Courses Page to Experimental Linguistics Laboratory to Dept. of Linguistics to University of Manitoba Navigation Bar

Welcome to the "Monthly" Mystery Spectrogram webzone! To access any of the other pages of the MMS webzone, or the 'host' pages of the lab, department, or university, you can use the new navigation grid (above) or the navigation links at the bottom of the page. Onward!

News for 2009: "Monthly" to be interpreted loosely

Well, a note of explanation is called for. First, the end of 2008 went badly—travel plans gone awry, old paperwork dragging on and on, new paperwork piling up. Then there was January, which also went badly. February and March were largely taken up with committee work, graduate committee stuff, and trying desperately to come up with a plan for my impending "research leave" (because apparently "sabbatical" is not a term-of-use up here). So the next thing you (I) knew, the end of term was upon me and I was still working on a January Spectrogram.

And no one seemed to notice. Certainly no one complained.

So I'm going to give myself a break, at least for the rest of 2009 and possibly for the duration of my leave, which ends on 30 June 2010. Until further notice, we should read "monthly" as "occasional", and understand "occasional" as "whenever Rob d*mn well pleases", at least as far as the "monthly mystery spectrogram" is concerned.

And now we come to what some of you presumably have been waiting at least four months for:

Mystery Spectrogram for Early 2009

Mystery Spectrogram for Early 2009

This is my voice (west coast American English with hints of mid-west and Canadian here and there). This is a semantically plausible (not to say 'predictable'), declarative sentence containing no proper names.

By the way, I'm playing with a new presentation. Since 'everyone' is moving to Praat for their analysis, I'm playing with it to see if I can produce mystery spectrograms with it. I didn't really like the bandwidth resolution of this presentation, or the preemph if it comes to it, but I did my best. To compensate for the cruddy formant resolution, I let it overlay a format track. Note how the LPCs are off, particularly in the lower frequencies. This is why I hate LPCs. But whatever.

About sound spectrograms

How do I read a spectrogram?

First, go read the acoustic phonetics chapter of the book of your choice, or work your way through my "How To" section for a quick overview of the procedure. Then as always, learn by doing!

Every month or so, I'll put up a new spectrogram for you to decipher. You can e-mail me guesses and questions, if you like. You can access the solution (with excruciating explanation) to the previous month's mystery using the "Solution" button in my navigation bar at the top of this page. For a few general hints, click on the "How To" icon, and for extensive practice, work through the mystery spectrogram "Past Mysteries" section.

What is a spectrogram?

A sound spectrogram (or sonogram) is a visual representation of an acoustic signal.

To make a spectrogram, a Fourier transform is applied to an acoustic wave (or more technically, its electronic analog), deriving the frequencies and amplitudes of its component simplex waves. A spectrogram represents that information visually. Time goes from left-to-right, with frequency represented vertically. Energy (loudness) is represented by darkness (in these kinds of displays--there are other "palettes" that are used for different applications). A pulse of the vocal folds produces a moment of high acoustic energy. Multiple pulses (as in normal voicing) come out on the spectrogram as closely spaced vertical striations. These striations are darker at some frequencies than others, and over time these darker bits form time-varying bars called "formants". Formant structure is a principal cue to vocal tract shape, and thus to articulatory configuration.

For more information, check out the "How To" section.

Why would I want to read one?

This activity is principally inspired by two things, both largely the influence of the late Peter Ladefoged. The first is the course at UCLA that all of us in the lab took or sat in on during our years there, "Practicum in Phonetic Data Analysis", which for years principally consisted of spectrogram reading. There's no better way to get used to thinking and 'imaging' the acoustics of speech (and indirectly the complexities of articulation and perception), relating physical acoustic events with underlying and surface phonological/phonetic models, and basically coming to terms with the whole segmental-prosodic 'problem' in continuous speech. The second is the unlabeled spectrograms in the acoustics chapter of A Course in Phonetics. Even if you've never been to a spectrogram reading session at UCLA, these have provided similar experiences to countless people around the world.

So that's the linguistic pedagogical aspect. There's a more general pedagogical aspect, in that the whole activity is practice at the scientific method. You make some observations, form some hypotheses, and then look for corroborating or disconfirming evidence.

That and it's just plain fun. If you're into phonetics at all, and have any affinity for crossword puzzles, cryptograms, or that sort of puzzle, decoding spectrograms is the activity for you. Believe it or not, I've heard from non-speech scientists who use this website as a coffee-break.

As it always has, The Monthly Mystery Spectrogram will continue to draw its inspiration from Peter Ladefoged. The Master, as I sometimes referred to him, passed away in January of 2005. I, among so many others, will miss him terribly. I hope that the Monthly Mystery Spectrogram website can be a small, continuing expression of gratitude to his generosity as mentor, teacher, and friend.