Spectral envelope

This video just has a plain transcript, not time-aligned to the videoIt's becoming very clear that the frequency domain is the best place to inspect speech sounds.
We're going to define a key property of speech sounds now called the 'spectral envelope' that can only be inspected in that domain.
Here come four speech sounds; we've heard them before.
The word 'but' - I'm just going to plot the vowel, but we're going to play the whole word.
The word 'bat' - again plot just the vowel, but we'll listen to the whole word.
And a couple of unvoiced sounds.
Comparing all four of these sounds, there are some really quite large differences.
The most obvious is that the amount of energy at different frequencies varies.
We can draw a line around these spectra, and that's called the spectral envelope.
Because the spectral envelope can take on so many different shapes, it can carry an awful lot of information.
We make many differences between spectral envelopes, and speakers do that.
Speakers manipulate the spectral envelope as part of conveying the message to the listener.
To reinforce that, let's look a whole sentence, 'Susan works at the zoo on Saturdays.'
Let's use short term analysis.
There's no point plotting the spectrum of the whole sentence, but we need to zoom in to small regions of it.
There's an analysis frame: that's just the silence at the beginning.
That's not very interesting.
I'm going to slide this analysis frame slowly through the utterance and look at the spectral envelope on the right while I'm doing that.
The spectral envelope is constantly changing: it's always on the move.
That's the speaker manipulating their vocal apparatus to change the spectral envelope to convey information to the listener.
We see very distinctive patterns in the voiced regions like this one, or the unvoiced regions.
So the next question will be, 'How is the speaker manipulating the spectral envelope?'
To understand that, we need to go back to some speech production.
We need to understand some physics: that the vocal tract is a tube and that tube has a property called 'resonance'.
We'll then talk about that in phonetic terms.
We'll talk about the vocal tract resonances, which are called 'formants'.
Then, we'll actually just generalise that in more engineering terms and talk about the vocal tract being a filter that places a spectral envelope on a very basic sound source, such as the impulse train that is playing the role of the vocal folds.