When we speak, our brains organize language into discrete categories—phonemes like /p/, /t/, or /k/. Yet the physical sound wave leaving our lips is a continuous, highly variable stream of air pressure changes. How do listeners reliably map this continuous signal back into discrete categories, and how do speakers maintain these distinctions?
The answer lies in quantitative acoustic-phonetic measurement. By extracting precise temporal, spectral, and source-based parameters from speech recordings, researchers can reverse-engineer the physical mechanics of the vocal tract. This approach reveals how abstract phonological contrasts are rooted in measurable, multi-dimensional acoustic landscapes.
The Temporal Domain: Timing and Duration
The most intuitive way to distinguish speech sounds is through time. When a speaker produces a stop consonant (like /p/ or /t/), the vocal tract completely blocks the airflow and then releases it in a sudden burst. Two primary temporal metrics capture this gesture: closure duration (how long the airway remains blocked) and Voice Onset Time (VOT) (the interval between the release of the closure and the onset of vocal fold vibration).
We might expect a single consonant and a doubled ("geminate") consonant to be distinguished solely by how long the speaker holds the closure. However, physical speech production is rarely that simple. In Phonetic geminates in cypriot greek: the case of voiceless plosives, Christiana Christodoulou demonstrates that in Cypriot Greek, VOT can serve as a more reliable cue than closure duration for distinguishing geminates, particularly at the beginning of words where closure onset is difficult to detect. Similarly, when investigating sonorant contrasts in Acoustic Cues to the Singleton-Geminate Contrast: The Case of Libyan Arabic Sonorants, Amel Issa confirms that while duration is the primary anchor, it operates alongside subtle adjustments in the surrounding vocalic context.
Temporal measurements also reveal how speech adapts under physical time pressure. When speakers accelerate, they do not just speed up uniformly; they compress sounds until distinct segments blur together. Chierh Cheng et al. analyzed this in Extreme reductions: contraction of disyllables into monosyllables in taiwan Mandarin, proving that the extreme contraction of two syllables into one is not a sudden, categorical phonological rule. Instead, it is a gradient "undershoot" of articulatory targets that occurs naturally when the available time drops below a critical physical threshold.
The Spectral Domain: Resonance and Vocal Tract Shape
While timing tells us when articulators move, the frequency spectrum tells us where they are. As sound waves travel from the vocal folds through the mouth, the vocal tract acts as an acoustic filter. It amplifies specific frequencies based on its shape, creating peaks in the spectrum known as formants.
The first two formants ($F_1$ and $F_2$) directly map to tongue position: $F_1$ correlates inversely with tongue height, while $F_2$ tracks tongue backness. When language learners map a new language onto their native phonetic space, these formant frequencies reveal exactly where their pronunciation goes astray. For instance, Akiyo Joto used formant measurements in The intelligibility of the English vowel /ʌ/ produced by native speakers of Japanese and its relations to the acoustic characteristics to show that when native Japanese speakers produce an unintelligible English /ʌ/, their tongue positions physically drift into the acoustic formant space of the Japanese /a/ or English /æ/.
Vocal Tract Shape ──> Acoustic Filter ──> Formant Frequencies (F1, F2)
(Tongue Height/Backness) (Quantifiable Spectral Peaks)
For consonants that lack a clean harmonic formant structure—such as fricatives like /s/ or /ʃ/—frication noise is analyzed using the Center of Gravity (COG). COG calculates the mathematical mean of the frequencies in a sound wave, weighted by their intensity. A smaller, more forward oral cavity (like the space in front of the tongue for /s/) filters out lower frequencies, driving the COG higher. A deeper constriction pushes the COG lower.
Jane Wottawa et al. leveraged COG and intensity bands in Putting German [ʃ] and [ç] in Two Different Boxes: Native German vs L2 German of French Learners to uncover a striking perceptual divide: while native German speakers rely on contextual vowel transitions to separate the similar fricatives [ʃ] and [ç], second-language learners depend almost entirely on the internal COG of the fricative noise itself to distinguish the sounds.
The Source Domain: Glottal Dynamics and Voice Quality
Beyond the shaping of the oral cavity, the sound source itself—the larynx—provides critical acoustic cues. The rate at which the vocal folds vibrate determines the fundamental frequency ($f_0$), or perceived pitch. However, the physical manner of that vibration introduces variations in voice quality, such as breathiness, creakiness, or hoarseness.
These source features often co-vary with consonant articulation. For example, when a speaker releases a voiceless consonant, the vocal folds take a moment to tense and stabilize, causing the pitch ($f_0$) at the start of the following vowel to be higher than it would be after a voiced consonant. Jasmin Pöhnlein et al. tracked this in The emergence of obstruent-intrinsic f0 and VOT as cues to the fortis/lenis contrast in West Central Bavarian, discovering that younger speakers are actively adopting this "obstruent-intrinsic $f_0$" as a primary cue to maintain the contrast between strong (fortis) and weak (lenis) consonants, capturing a sound change in progress.
To quantify more subtle voice qualities, researchers measure micro-perturbations in the speech signal: jitter (cycle-to-cycle variations in frequency) and shimmer (cycle-to-cycle variations in amplitude). These metrics allow phoneticians to isolate pragmatic or physiological shifts in speech. For example, Kikuo Maekawa et al. demonstrated in Voice-Quality Difference Between the Vowels in Filled Pauses and Ordinary Lexical Items that the vowels in Japanese filled pauses (like "um") exhibit distinct jitter and shimmer profiles compared to ordinary words, meaning speakers unconsciously alter their voice quality to signal a hesitation.
On a clinical level, tracking these exact parameters offers an objective window into physical recovery. In Speech quality after major surgery of the oral cavity and oropharynx with microvascular soft tissue reconstruction, Irma M. Verdonck-de Leeuw et al. showed that post-operative changes in jitter, shimmer, and the harmonics-to-noise ratio (HNR) precisely mirror the structural degradation and subsequent rehabilitation of a patient's speech mechanisms.
Why It Works: The Multiplicity of Cues
The core reason acoustic feature extraction is such a powerful framework is that human speech relies on a principle of redundancy. A single linguistic contrast is rarely left to depend on a single acoustic parameter. Instead, speakers deploy a multiplicity of covarying cues to ensure that their message survives the noise of the real world.
As Míša Hejná observed in Multiplicity of the Acoustic Correlates of the Fortis-Lenis Contrast: Plosives in Aberystwyth English, a phonological distinction like fortis vs. lenis is supported by an entire matrix of features working in tandem: release duration, preceding vowel duration, pre-aspiration, and localized breathiness.
If one cue is obscured or altered by coarticulation—the physical blending together of adjacent sounds studied by Jared Sharp et al. in Coarticulation of Sibe Vowels and Dorsal Fricatives in Spontaneous Speech: An Acoustic Study—the remaining cues in the acoustic matrix shift to preserve the contrast. By extracting and analyzing these multi-dimensional parameters, researchers can look past the surface-level variability of spoken audio to map the underlying, invariant structures of human language.
Key Foundational Work
An acoustic-analytic role for the deviation between the scansion and reading of poems (2004) — contributed 1 concept elements to this direction.
Towards a new level of anotation detail of multilingual speech corpora (2004) — contributed 1 concept elements to this direction.
Modelling and ranking of differences across formants of british, australian and american accents (2004) — contributed 1 concept elements to this direction.