Skip to content

LESSON 42

How to read a spectrogram

A spectrogram is a picture of speech that shows time, frequency and intensity together. In this lesson we cover the spectrogram axes, wideband and narrowband spectrograms, the F1 and F2 formants of vowels, the high-frequency energy of fricatives, voice onset time (VOT) in stops, and why all this matters in high-frequency hearing loss.

Watch on YouTube

Follow along

Hello, and welcome to the AIHEAR Audiology Academy.
In this lesson we will learn to read the spectrogram, the picture of speech.
We will look at vowel formants, the high-frequency energy of fricatives and voice onset time in stop consonants.
We can show a sound in different ways; the first is the waveform, which shows pressure changing over time.
The spectrum shows how much energy a sound carries at each frequency at a given moment.
The spectrogram combines the two: it shows how the frequency content changes over time.
Because speech is a constantly changing sound, the spectrogram is one of the most powerful tools for studying it.
The first sound spectrographs were developed at Bell Laboratories in the 1940s and transformed speech research.
On a spectrogram, the horizontal axis is time; speech runs from left to right.
The vertical axis is frequency, with low frequencies at the bottom and high frequencies at the top.
The third dimension, intensity, is shown by darkness or color; dark or bright areas mean more energy.
Note that frequency runs from left to right on the audiogram, but from bottom to top on the spectrogram.
With these three axes in mind, we can read the traces of speech sounds.
Spectrograms can be made with two basic settings: wideband and narrowband.
In a wideband spectrogram the analysis window is short, and formants appear as dark bands.
Time detail is good; each vibration of the vocal folds can be seen as a thin vertical line.
In a narrowband spectrogram the window is long; this time the individual harmonics appear as horizontal lines.
Frequency detail is good, but fast changes over time become blurred.
Both are valid: wideband is usually chosen for formants and consonants, narrowband for pitch and harmonics.
To understand vowels, we use the source-filter model.
The source is the vibration of the vocal folds; its rate is called the fundamental frequency, or F0.
F0 is about one hundred to one hundred fifty hertz in men, two hundred to two hundred fifty in women, and higher in children.
Along with the fundamental, the vocal folds produce its multiples, the harmonics; the filter is the vocal tract from the larynx to the lips.
The vocal tract boosts certain frequency regions; these resonance regions are called formants.
As the tongue, jaw and lips change position, the formants move and different vowels emerge.
Each vowel is recognized largely by the position of its first two formants, F1 and F2.
The first formant is inversely related to tongue height: in high vowels such as /i/ and /u/, F1 is low.
In low vowels with an open mouth, such as /a/, F1 is high, around seven hundred hertz.
The second formant relates to tongue frontness: F2 is high in front vowels such as /i/ and /e/, and low in back vowels such as /u/ and /o/.
The values in the table are approximate examples for an adult male speaker; formants are higher in women and children.
Each of the eight Turkish vowels sits in its own place on the map formed by these two formants.
Vowels are the strongest sounds of speech, and most of their energy lies below one thousand hertz.
That is why even people with high-frequency hearing loss usually hear vowels well.
But telling vowels apart also needs the second formant; in front vowels such as /i/, this formant rises above two thousand hertz.
With mid and high-frequency loss, Turkish vowel pairs such as i and ü, or e and ö, may be confused.
On the speech banana, too, we see vowels gathered in the low and strong region, and consonants in the higher and weaker region.
Fricatives are produced by turbulence as air is forced through a narrow gap in the mouth.
On the spectrogram they appear not as regular harmonics but as grainy noise spread over a wide area.
The energy of the s sound lies in the highest region, usually concentrated above four thousand hertz.
For sh, the space in front of the constriction is longer and the lips are slightly rounded.
So the energy of sh starts lower, at around two thousand to two thousand five hundred hertz.
The f sound is a weak noise spread over a wide area, which makes it one of the most easily missed sounds.
In voiced fricatives such as z, v and zh, besides the noise, a voicing bar from the vocal folds is seen at the very bottom.
In stop consonants, the airflow is completely blocked for a moment; on the spectrogram this closure appears as a short silent gap.
Then, as the closure is released, a short vertical burst is seen.
The time between the burst and the start of vocal fold vibration is called voice onset time, or VOT.
In Turkish voiced stops such as b, d and g, voicing often starts with the burst or even before it.
In voiceless stops such as p, t and k, voicing starts with a clear delay, after a breathy aspiration noise.
This difference of a few tens of milliseconds shows how precisely the brain uses timing information.
This example audiogram shows a loss that drops steeply above one thousand hertz; most vowel energy stays in the audible region.
But the energy of fricatives such as s, sh and f lies above the threshold line, in the inaudible region.
The person hears that someone is talking, but may misunderstand the words.
In Turkish, many suffixes and words contain the s sound; missing it can change the meaning.
With female and child talkers, s energy shifts even higher; this explains why wide-bandwidth amplification matters for children.
A practical use of spectrogram knowledge is the Ling six sound test.
It uses the sounds m, oo, ah, ee, sh and s, which together roughly cover the frequency range of speech.
M and oo represent the low region, and ah the mid region.
Ee carries a low first formant and a high second formant, while sh and s represent the high region.
It is used, especially with children, to quickly check whether the device passes these sounds; we will cover it in detail later.
When reading a spectrogram, first look for silent gaps; these are often the closure phase of stops.
Then look for the voicing bar at the bottom and the thin vertical lines showing vocal fold vibration.
In vowels, find the dark formant bands and note where F1 and F2 lie.
Wide, grainy noise areas in the high region point to fricatives.
Finally, look at formant transitions between consonant and vowel, and at the duration of the sounds.
You can try these steps by making a spectrogram of your own voice with free programs such as Praat.
A tip from the clinic: in speech testing, note not only how many answers are correct, but also which sounds the patient confuses.
Confusions of s, sh and f are often linked to the region of the audiogram between two thousand and eight thousand hertz.
For example, a patient who hears sh or f instead of s may lack enough audibility in the high region.
These notes provide valuable information for device fitting and counselling.
On a spectrogram, the horizontal axis shows time, the vertical axis frequency, and darkness shows intensity.
Vowels are recognized by the position of the first two formants, the resonances of the vocal tract.
Fricatives appear as noise in the high region; the energy of s is usually above four thousand hertz.
In stops, the closure gap, the burst and voice onset time separate voiced from voiceless sounds.
Test yourself: which sound appears as noise in the highest region of a spectrogram?
The correct answer is C: the consonant s.
S is a fricative, and its energy usually appears above four thousand hertz as grainy noise.
In the next lesson we will place speech sounds on the audiogram and look at the speech banana.
Subscribe to the channel so you never miss a new lesson. See you next time!

Keywords

Click a keyword to search all lessons for that term.

Prerequisites

LESSON 39 Pure Tones vs Complex Sounds: Fourier Basics Coming soon

Comments

Questions and comments appear after approval.

No comments yet. Be the first to comment.