Two people read the same sentence aloud. A speech-to-text system writes different words for one speaker even though listeners understand both clearly. A third person concludes that one accent is less accurate. That conclusion confuses a technological recognition error, a social judgement and an acoustic observation. The gold standard of phonetics is to describe speech sounds with the right evidence, while recognising that the sounds of language vary across speakers and contexts.
Phonetics studies how speech is produced, transmitted and perceived. Its main traditions are articulatory phonetics, acoustic phonetics and auditory or perceptual phonetics. It connects linguistics, language teaching, speech technology, language documentation and communication research. A description of how someone pronounces a word is not automatically a diagnosis, an assessment of intelligence or a claim that one accent is superior.
This O–P root complements The Gold Standard of Linguistics, Sociolinguistics, Language Documentation and Pedagogy. It focuses specifically on identifying speech phenomena, measuring them and drawing proportionate conclusions.
Quick information: what makes a phonetic explanation reliable?
A dependable analysis identifies the language or variety, speakers and task, states whether it concerns a sound category or a particular pronunciation, uses transcription suited to that purpose, and checks any acoustic measurement against actual audio. It reports recording conditions, the position of segment boundaries, relevant variability and uncertainty. It does not turn dialect difference into deficit.
The International Phonetic Association maintains the official International Phonetic Alphabet and chart. The chart symbolises articulatory and other phonetic distinctions; it does not say every language uses every sound or that one symbol alone predicts the exact production of every speaker. The Association’s Illustrations of the IPA show how language-specific descriptions can be accompanied by transcriptions and speech recordings.
Find your route
- Quick information: what makes a phonetic explanation reliable?
- Find your route
- Worked fictional example: why ‘pin’ and ‘spin’ sound different
- Eighteen controls for a strong phonetic study
- Second worked case: a higher pitch mean is not proof of a different tone
- A phonetics quality-review checklist
- Frequently asked questions about phonetics
- Continue through the eduKate language and education ecosystem
- Sources and responsible scope
Worked fictional example: why ‘pin’ and ‘spin’ sound different
In one common analysis of English, the word ‘pin’ contains the phoneme /p/. At the beginning of a stressed syllable, many English varieties produce this stop with strong aspiration, which may be transcribed narrowly as [pʰ]. In ‘spin’, the /p/ following /s/ is often produced with much less aspiration. The spelling letter p is the same, but the phonetic realisations can differ by position.
Suppose a researcher records a fictional speaker producing ten tokens of ‘pin’ and ten tokens of ‘spin’. An analysis gives a mean voice onset time of 55 milliseconds for the initial stop in ‘pin’ and 10 milliseconds for the stop in ‘spin’. These are invented values for teaching, not universal English norms. Voice onset time here is measured from a stop release to the onset of periodic voicing in the following sound, with the analyst checking whether a valid interval can be identified.
The result supports a bounded claim about those productions under those recording conditions. It does not show that the speaker pronounced one word correctly and the other incorrectly. It does not automatically establish that all speakers of the same variety have the same timing, or that the two contexts should be treated as independent of surrounding phonetic material.
A clearer study specifies stress, speech rate, phrase position, speaker, microphone, segmentation rules and whether ambiguous tokens were retained. The difference between phonemic notation /p/ and a narrower representation such as [pʰ] matters because phonetic details and contrastive linguistic categories are different questions.
Eighteen controls for a strong phonetic study
01. Ask a question about speech that can be heard or measured
Replace ‘Why is this accent bad?’ with a legitimate inquiry: How does vowel duration vary with speech rate? How are final consonants released? Which cues distinguish a contrast for listeners in this community? Avoid leading questions that assume one variety is defective.
Specify whether the unit of interest is a token, speaker, word, sentence or community. The purpose determines what evidence is needed.
02. Identify the language, variety and speaking task
Read speech, spontaneous conversation, imitation and careful citation forms can differ. Label the language variety responsibly rather than relying on a broad national category. Speakers may shift style across tasks or audiences.
Do not treat one reader as representative of an entire language. State sampling criteria and the conditions under which the data were collected.
03. Separate phonetics, phonology and orthography
Phonetics concerns actual speech production, acoustics and perception. Phonology concerns the organisation and function of sound patterns in a language. Orthography concerns writing conventions. A single letter can correspond to multiple sounds; multiple spellings can represent one sound.
Make the question explicit. A phonetic description may be narrow and detailed; a phonological transcription selects the contrasts relevant to a particular system.
04. Use the IPA for an identified purpose
The International Phonetic Alphabet is a tool for systematic sound description, not a universal replacement for ordinary spelling. The official IPA chart includes consonant, vowel, diacritic and prosodic notation.
State whether slashes represent a phonemic transcription and square brackets a phonetic transcription in the conventions being used. Explain diacritics when they carry a relevant distinction, and avoid adding impressive-looking detail unsupported by the recording.
05. Describe consonants using a reproducible feature system
For a pulmonic consonant, place of articulation, manner of articulation and voicing are useful starting dimensions. The sounds [p], [t] and [k] differ in place; [p] and [b] are often contrasted as voiceless and voiced in broad phonetic descriptions, though the actual cues in languages are more complex than labels alone.
Check actual production and language context. Descriptive labels are hypotheses to be tested against data, not substitutes for listening.
06. Describe vowels as speech categories with continuous variation
Vowel quality relates to articulation and acoustics, often represented through height, backness and rounding. Vowels also vary in duration, movement and context. A vowel chart locates categories; it does not show every actual trajectory a tongue takes.
Acoustic analyses commonly use formant estimates, but those estimates can be unstable depending on recording, speaker, pitch, nasalisation and analysis settings. Inspect the data rather than accepting every automated number.
07. Distinguish stress, rhythm, tone and intonation
A stressed syllable may be prominent through several interacting acoustic cues. Lexical tone can distinguish words in some languages, while intonation can express phrase-level structure, discourse functions or stance. These categories should not be merged under the vague label ‘voice goes up’.
Define the linguistic function and measurement. A pitch contour on its own does not tell the researcher whether a difference is lexical, grammatical or pragmatic.
08. Preserve clean, consented recordings
Record consent and intended reuse, especially when collecting personal voices. Document microphone, sampling rate, distance, background noise, room conditions and file handling. Preserve unedited originals where lawful and appropriate.
A transcript alone cannot recover the timing and spectral detail of lost audio. Conversely, high-fidelity audio does not justify indefinite disclosure of a participant’s identifiable voice.
09. Use waveforms and spectrograms as evidence, not decoration
Waveforms show signal amplitude over time; spectrograms display how acoustic energy is distributed across frequency and time under selected analysis settings. They can help locate bursts, frication, formants and voicing cues.
The Praat software documentation describes spectrogram, pitch, formant and segmentation functions. A graph is not self-interpreting: window length, frequency range, filtering and display settings alter what appears clear.
10. Define where a sound begins and ends
Segment boundaries are often ambiguous because speech is continuous and articulatory gestures overlap. A release burst or onset of voicing may be easier to identify than a boundary inside a smoothly changing vowel.
Write annotation rules, mark uncertain intervals, and test consistency across annotators where feasible. Treat a software-generated boundary as a proposal until it is checked against speech and context.
11. Distinguish voice onset time from a general ‘speed’ claim
Voice onset time is a specific relation between a stop release and onset of voicing, under an operational convention. It is not the duration of a word or the time needed to pronounce a letter. Comparisons require similar segment contexts and a defensible release point.
Report whether measurements are positive, negative or excluded by the chosen convention. Do not assume every stop token yields a clean measurable value.
12. Evaluate formants and fundamental frequency cautiously
Formants relate to resonant properties of the vocal tract, and fundamental frequency relates to periodicity when it can be estimated. Automated tracking can produce octave errors, spurious candidates and inconsistent formant identities.
Check measurements against the signal and appropriate settings. Compare like with like: a high mean pitch does not establish emotion, identity, sincerity or illness.
13. Consider coarticulation and adjacent sounds
A sound changes with neighbouring sounds because speech organs prepare and overlap gestures. Vowels can differ before nasal consonants; consonants can assimilate features across boundaries. That does not automatically create a new phoneme.
Record the surrounding sounds and compare within suitable contexts. A contrast measured in different word positions may reflect context rather than a property of the segment alone.
14. Sample speakers rather than multiplying tokens from one person
Twenty readings by one speaker describe within-speaker behaviour; they do not count as twenty independent speakers. Language-level generalisations require appropriate speaker sampling and an analysis that respects repeated observations.
State age range or other sample properties only where justified, consented and relevant. Avoid turning aggregate acoustic tendencies into rigid rules about an individual.
15. Validate perception separately from production
A measurable acoustic distinction does not automatically mean listeners reliably hear it as a contrast. Use suitable identification, discrimination or comprehension tasks when the question concerns perception, and describe the stimulus set and listener population.
Likewise, listeners may use multiple cues. An automated recogniser’s mistake is not proof that a human speaker produced an incorrect sound.
16. Compare accents without ranking human worth
A phonetic description can compare vowels, rhoticity, stress and consonant timing across varieties. It cannot establish that the users of one variety are intrinsically more intelligent, honest or articulate. Standardisation in a teaching or broadcasting setting is a social practice, not evidence of universal linguistic superiority.
Distinguish a target variety required for a particular task from claims about legitimate home and community varieties. Include speakers in decisions about how their speech is represented.
17. State the limits of automatic speech technology
Speech recognition, speaker embeddings, formant trackers and pitch algorithms can fail under background noise, unfamiliar accents or data imbalance. Compare outputs against representative human-reviewed material, not merely an aggregate vendor claim.
Protect recording consent, privacy and retention. Do not infer a sensitive personal attribute, health condition or identity merely from a speech recording and a model score.
18. Deliver a traceable phonetic description
A credible report can identify the recording set, permission scope, relevant language variety, transcription convention, annotation rules, analysis settings, sample sizes, uncertainty and audio-linked examples where reuse is lawful.
The gold standard is a description that other analysts can inspect, challenge and refine. It is not a claim that one pronunciation or one analysis tool is compulsory for every language and setting.
Second worked case: a higher pitch mean is not proof of a different tone
Suppose a fictional study compares eight tokens produced in a fast-speaking task with eight in a slow-speaking task. The fast tokens have a mean measured fundamental frequency of 180 Hz and the slow tokens 165 Hz. The difference is 15 Hz, but the word, phrase position, speakers and emotional context may differ, and both sets could include pitch-tracking errors.
It would be incorrect to announce a different lexical tone solely from the difference in means. Researchers should compare contextualised pitch trajectories and perceptual function, report the relevant language system, and account for within-speaker repetition and uncertainty. If the tokens all come from one speaker, the study does not thereby include sixteen independent speakers. All figures here are illustrative.
A phonetics quality-review checklist
- What language variety and kind of speech were recorded?
- Did participants consent to recording, analysis and any publication?
- Are transcripts labelled as phonemic or phonetic at an appropriate level of detail?
- Are place, manner, vowel, timing and prosodic descriptions grounded in actual audio?
- Were microphone and software settings recorded?
- Are segment boundaries defined and ambiguous items handled transparently?
- Is the experimental unit a sound token, word, recording or speaker?
- Were acoustic summaries checked for pitch and formant-tracking errors?
- Are perception findings supported by listener data rather than assumed from production?
- Does the conclusion avoid turning accent difference into personal deficiency?
Frequently asked questions about phonetics
What is phonetics in simple language?
Phonetics is the study of the physical production, acoustic properties and perception of speech sounds.
Is phonetics the same as phonology?
No. The subjects overlap, but phonology generally examines how sound patterns function within linguistic systems, whereas phonetics examines the actual sounds and their production and perception.
What is the International Phonetic Alphabet?
The IPA is an internationally used notation system for describing speech sounds. Its chart is maintained by the International Phonetic Association.
Why does IPA transcription use both slashes and square brackets?
In common linguistic practice, slashes indicate phonemic representations and square brackets phonetic realisations. The level of detail and assumptions should be explained for the particular analysis.
Can software determine a speaker’s accent perfectly?
No. Automatic models have limitations and can reflect biases or fail under unfamiliar recording conditions. A good evaluation uses representative and consented data.
Does a word always sound the same when repeated?
No. Rate, stress, adjacent sounds, context and individual production can change details even when listeners recognise the same word.
Does a non-standard accent mean a student speaks incorrectly?
No. Languages contain systematic regional and social varieties. A specific educational task can set a pronunciation target without treating another community variety as deficient.
How can a student begin learning phonetics?
Listen to a short recording, compare a few sounds, learn relevant IPA symbols, describe place and manner, and test the observations against more examples before generalising.
Continue through the eduKate language and education ecosystem
- The Gold Standard of Linguistics — language structure and linguistic explanation.
- The Gold Standard of Sociolinguistics — language variation, society and responsible interpretation.
- The Gold Standard of Language Documentation — documenting languages with community involvement and valid evidence.
- The Gold Standard of Historical Linguistics — sound change and language history.
- The Gold Standard of Computational Linguistics — tools, corpora and language models.
- The Gold Standard of Natural Language Processing — language technologies and their evaluation.
- The Gold Standard of Semantics — meaning and interpretation.
- The Gold Standard of Pedagogy — careful teaching decisions and evidence of learning.
- The Gold Standard of Education — the wider learning and subject ecosystem.
Sources and responsible scope
Authoritative starting points include the International Phonetic Association’s official chart, the IPA’s Journal and Illustrations descriptions, the IPA recording resources, and the Praat phonetics software and documentation. Read the actual language documentation and check recordings before generalising. The examples and measurements in this article are fictional and are not population norms.
This article is educational, not a clinical speech assessment, a prescribed pronunciation standard or an automatic speaker-identification system. Return to The Gold Standard and its Sectors and Services Directory.