Prosody is the organisation of speech through patterns such as pitch, prominence, rhythm, timing and phrasing. It helps listeners hear which information matters, how words belong together and what a speaker is doing with an utterance. The gold standard of prosody is an explanation that connects a heard pattern to evidence and context, then checks whether that explanation still works when the speaker, sentence or listening situation changes.
You can begin without specialist equipment. Compare two readings of the same sentence, identify what changes, and ask a listener what each version suggests. For measurement, keep the recording, the analysis settings and the meaning claim separate. A rising line on a pitch display is an acoustic observation; a conclusion about uncertainty, politeness or a question requires further evidence. Every numerical dataset and dialogue in this guide is an original hypothetical teaching example, not a reported experiment or a norm for judging people.
Use the Gold Standard master to explore related subjects, the Education parent to examine learning that transfers, and How X Works to follow mechanisms across fields. This guide concentrates on spoken prosody. It does not provide a universal dictionary of tones or a method for inferring personality from a voice.
Choose your prosody route
Chapters 1–3: Hear structure, sound and prominence
Chapters 4–6: Understand phrasing and measure pitch
Chapters 7–9: Check recordings, rates and listening
Chapters 10–12: Annotate, repair and test transfer
Chapter index
Hear structure, sound and prominence · Chapters 1–3
1. Hear the organisation beyond the individual words
Understand phrasing and measure pitch · Chapters 4–6
4. Treat phrasing as a grouping problem
Check recordings, rates and listening · Chapters 7–9
7. Check the pitch track before interpreting it
8. Distinguish speaking rate from time spent articulating
9. Design a listening comparison that can answer its question
Annotate, repair and test transfer · Chapters 10–12
10. Annotate disagreements instead of hiding them
Answer questions and continue learning · Chapters 13–14
13. Answer common questions without overstating the evidence
14. Build a learning route that ends in a checkable explanation
CHAPTER 1 OF 14 · Hear structure, sound and prominence
1. Hear the organisation beyond the individual words
Imagine the sentence, “Mira sent the revised diagram.” It can answer who sent something, what Mira sent, or whether the diagram was revised. The words remain the same while the prominence and surrounding conversation change. A listener uses the whole event, including what was asked beforehand, to interpret the answer.
This is why a word-perfect transcript is not a complete record of speech. It preserves lexical material but may omit which syllable was prominent, whether the speaker continued without a break, or how a phrase ended. Conversely, a pitch trace without words does not tell you what proposition was expressed. A useful account connects the layers.
For deeper explanations of the two sides of the exchange, use prosodic encoding for how speakers organise delivery and prosodic parsing for how listeners recover grouping. This root connects those mechanisms with measurement, research design and practical checks.
Ask what information the pattern supplies
Start with a question that could be answered by observation. Did a pause make two groups of instructions easier to distinguish? Did prominence help a listener identify the corrected colour? Did a phrase ending encourage someone to treat the turn as complete? Each question identifies an outcome that can be examined.
Avoid starting with a verdict such as confident, rude or intelligent. Those labels combine interpretation, expectations and social judgement. They may also conceal variation in speaking style, language background, task demands or recording quality. Describe the pattern first and state alternative explanations before drawing conclusions about the interaction.
The Max Planck Institute’s prosody research forum describes research on rhythm, melody and timing across languages, modalities and functions. This breadth is a useful reminder that prosody is a field of investigation rather than a set of presentation tips. Here we use practical spoken examples to make its central questions accessible.
For a first exercise, record yourself reading a short direction in two ways. Keep the wording fixed and change only the intended grouping. Before listening back, write what distinction you intended. Then compare that intention with what another listener actually understood. The difference between intended and received meaning is often the most useful finding.
CHAPTER 2 OF 14 · Hear structure, sound and prominence
2. Keep sound, perception and linguistic function distinct
Fundamental frequency, often written F0, describes the repetition rate of a periodic signal. During voiced speech it is closely connected with vocal-fold vibration. Pitch is a perceptual experience related to frequency, but the terms are not interchangeable in every situation. A program estimates a signal property; a listener experiences a sound in context.
Duration measures time. Intensity concerns acoustic energy, while perceived loudness depends on more than one physical quantity and on listening conditions. Prominence is broader still: a syllable may stand out through a combination of pitch movement, duration, intensity, vowel quality and its position in the phrase. It is therefore unsafe to identify prominence by selecting the tallest point on one graph.
Separate three entries in your notebook
For a fictional recording, write an acoustic observation: the selected vowel lasts 0.18 seconds. Then write a perceptual observation: two listeners heard that word as prominent. Finally write an interpretation: in this context the prominence appears to correct a previously mentioned destination. These claims have different evidence requirements.
If the duration measurement changes after boundary correction, that does not automatically overturn the listener judgements. If listeners disagree about the correction, that does not necessarily mean the time measurement is wrong. Keeping layers separate helps you locate the actual disagreement instead of treating the entire analysis as right or wrong.
A familiar word such as tone can cause confusion because it has several everyday and technical uses. It can refer loosely to an attitude, to pitch, or to language-specific contrasts. Define the intended meaning when precision matters. “The final pitch rises” is clearer than “the tone is positive” when the evidence consists of an acoustic contour.
The same discipline applies to stress. A word’s lexical stress pattern and its prominence in a particular sentence are related but different questions. A learner may know how a word is normally stressed yet still need to decide which word should stand out in a correction. Teaching both questions as simply “say it louder” hides the distinction that the learner needs.
CHAPTER 3 OF 14 · Hear structure, sound and prominence
3. Use prominence to identify the question being answered
Consider a fictional exchange. A colleague asks, “Who moved the meeting?” The answer is, “Ravi moved the meeting.” Prominence on Ravi can help present the person as the informative answer. If the question instead concerns what Ravi moved, prominence may fall elsewhere. The context supplies the alternatives against which the answer is understood.
This is not a rule that every sentence has one fixed focus or that all speakers realise it identically. It is a practical way to test an interpretation. Ask which question makes the utterance a relevant answer, then inspect whether its delivery supports that reading. If several questions fit, retain the ambiguity.
Build a correction that a listener can verify
Suppose someone says, “We need the green folder.” The correction is, “We need the blue folder.” A speaker may make blue prominent to distinguish the intended alternative. The useful outcome is whether the listener selects the blue folder, not whether the speaker produced a large pitch excursion.
Now change the correction to “We need the blue binder.” The competing alternatives have changed: the object category matters as well as the colour. A learner who simply stresses the first adjective in every instruction may preserve the wrong contrast. Ask them to explain what misunderstanding the new version repairs before practising its delivery.
The information-structure translation guide examines how topic, focus and alternatives affect a message across languages. The connection is practical: carrying over the words can leave behind the point of the correction. An appropriate target-language construction may express the distinction differently.
A useful classroom exercise has three roles: speaker, listener and observer. The speaker receives an intended correction, the listener chooses among labelled objects, and the observer records both the utterance and the selection. Rotate roles and change the objects. If the listener succeeds only when they already know the answer, the task has not demonstrated that the prosody conveyed it.
Do not require a caricatured voice to make the contrast visible. Naturalness and intelligibility should be considered alongside the intended distinction. The aim is control over meaning, including the ability to use explicit wording when prosody alone would be unreliable.
CHAPTER 4 OF 14 · Understand phrasing and measure pitch
4. Treat phrasing as a grouping problem
Speech arrives in time, and listeners must decide which words belong together. Prosodic phrasing can help organise that sequence. Pauses are one possible cue, but phrasing is not simply the presence or absence of silence. Timing and pitch patterns can contribute to a perceived boundary even without a long gap.
Imagine a spoken packing instruction: “Take the small boxes and the labelled envelopes.” A listener needs to know what small modifies and which objects belong to the requested set. Delivery can help, but it cannot reliably rescue a fundamentally unclear instruction in every context. If the grouping matters, revise the wording as well as the performance.
Repair a practical ambiguity
Use this fictional instruction: “After checking the label send the parcel.” A speaker can group the initial instruction separately from the main action. In writing, punctuation can help readers recognise the structure. In a noisy room, however, a subtle boundary may be lost. “First check the label. Then send the parcel” makes the sequence more explicit.
The revised instruction has a testable consequence: the recipient should check before sending. Asking whether it sounds better is useful for style but insufficient for checking the action sequence. Have a listener explain the order without seeing the text. If they reverse it, inspect both wording and delivery.
Different annotation systems represent boundaries differently. A beginner can start with clearly defined descriptive marks for an audible pause and a perceived phrase boundary, keeping those marks separate. Do not use one slash to mean silence, syntax, breathing and completion without explaining which interpretation is intended.
Boundary placement should also be considered across a whole message. A list read with identical continuation patterns may sound unfinished at its end. A sequence of short sentences with strong endings may sound disconnected. Experiment with a complete instruction, including its final item, rather than evaluating isolated clauses that never occur alone in the real task.
When assessing a learner, preserve the successful grouping they already use. Repair the specific point where the listener loses the structure. Asking for a more expressive reading without identifying the missing relationship leaves the learner guessing about which change will help.
CHAPTER 5 OF 14 · Understand phrasing and measure pitch
5. Examine intonation without universal rise-and-fall rules
Intonation concerns pitch patterns associated with utterances and phrases. A rise or fall can contribute to interpretation, but its function depends on language variety, construction, context and interaction. “Rising pitch means a question” is too simple to serve as an analysis of every utterance.
A person may use a question to request information, seek confirmation, challenge a claim or repair something they did not hear. Those actions need not share one contour. Similarly, a sequence that looks like a statement on paper may be used to check information. Start with the activity and the actual recording.
Compare two plausible interpretations
In a fictional exchange, one speaker says, “The train leaves at six.” The other repeats, “At six?” The repetition could request confirmation of the time or express surprise about it. To distinguish these possibilities, examine what happened before and what the participants do next. The punctuation supplied by a transcriber is already an interpretation.
Write down an alternative account before assigning a final label. If both confirmation and surprise remain possible, report that the evidence does not distinguish them. A useful analysis does not become stronger by eliminating uncertainty from the prose while leaving it unresolved in the data.
The original ToBI labelling guidelines provide a formal route into transcribing intonation and other prosodic properties. Their stated coverage is limited to particular English varieties; the authors explicitly reject treating the system as automatically suitable for every language or dialect. That scope restriction is part of the method, not an inconvenience to ignore.
For everyday learning, begin with listening and contextual interpretation before memorising labels. A notation becomes useful when it captures a distinction you can explain and locate in the sound. Copying a symbol sequence without understanding its relationship to the recording creates apparent precision without a dependable analysis.
Keep accent difference separate from communication failure. A familiar contour is not inherently more correct than an unfamiliar one. If a task requires a particular distinction, assess whether that distinction reaches the listener in the relevant setting, and explain the criterion openly.
CHAPTER 6 OF 14 · Understand phrasing and measure pitch
6. Measure pitch changes using an explicit reference
Suppose a hypothetical voiced interval has an estimated F0 of 120 hertz at one selected point and 180 hertz at another. The absolute difference is 60 hertz. The ratio is 180/120 = 1.5. These are two descriptions of the same pair of values, and they answer different comparison questions.
A second speaker changing from 200 to 260 hertz also changes by 60 hertz, but the ratio is 1.3. Equal differences on a hertz scale do not imply equal frequency ratios. When comparing speakers, decide which property your question requires and explain the reference used.
Work an original semitone calculation
Using the logarithmic relationship documented in Praat’s semitone conversion, a conventional interval measure is 12 × log2(frequency ratio). For the 120-to-180 example, this is 12 × log2(1.5), approximately 7.02 semitones. For 200 to 260, it is 12 × log2(1.3), approximately 4.54 semitones. A doubling gives twelve semitones because log2(2) equals one.
These calculations describe frequency relationships. They do not establish how surprised, polite or certain either speaker sounded. A researcher still needs a linguistic question, trustworthy measurements and evidence about perception or use. A mathematically correct transformation cannot supply a missing interpretation.
State whether the reference is a particular point, a baseline calculated from a recording, or some other value. Changing the reference changes the reported interval. If you use a speaker-specific baseline, explain how it was estimated and which parts of the recording were eligible. Do not silently include unvoiced gaps as zero-frequency observations.
For a reproducible notebook, retain the original hertz values alongside transformed values. Record the selected times and the reason for choosing them. A reader should be able to reconstruct the calculation and assess whether the selected points represent the feature you claim to measure.
Avoid reporting many decimal places merely because software provides them. Boundary uncertainty, recording conditions and pitch-tracking errors can dominate small numerical differences. The right precision is the precision justified by the method, not the longest number that fits in a spreadsheet cell.
CHAPTER 7 OF 14 · Check recordings, rates and listening
7. Check the pitch track before interpreting it
Pitch-tracking software estimates periodicity from an acoustic signal. It can make errors, particularly when the signal is noisy, irregular, weak or affected by unsuitable settings. A smooth coloured line is a computed result, not direct access to a speaker’s intention.
The Praat manual on choosing a pitch-analysis method explains method choices and examples of octave-related errors. It also identifies filtered autocorrelation as a preferred option for intonation and vocal-fold-vibration analysis. Use the documentation for the actual method, then inspect whether the resulting track fits the recording.
Diagnose an apparent jump
Imagine a short vowel whose displayed track jumps from about 150 to about 300 hertz and immediately returns. The doubling suggests a possible tracking error, but it is not proof. Listen, inspect the relevant signal and check the analysis settings before accepting or correcting the point. Never remove a value merely because it conflicts with the desired story.
The pitch-contour configuration guide explains how pitch-range settings affect analysis. An unsuitable floor or upper limit can exclude plausible values or encourage an inappropriate candidate. Retain the settings with the result so another analyst can reproduce the calculation.
A missing track does not necessarily indicate silence. Unvoiced speech can carry important words or parts of words without providing the same periodic source used for F0 estimation. Do not draw a line through every gap and then treat the invented values as measured pitch.
Maintain a correction record. For each changed or excluded interval, note the original output, the reason for intervention and the resulting decision. If uncertainty remains, mark the interval unusable for that particular measurement. A smaller transparent dataset is preferable to a complete-looking contour assembled from undocumented guesses.
Keep the research question proportionate to the recording. A compressed voice message may be adequate for a broad listening comparison while unsuitable for a fine acoustic claim. Choosing an analysis method begins with understanding what the evidence can support.
CHAPTER 8 OF 14 · Check recordings, rates and listening
8. Distinguish speaking rate from time spent articulating
A message can feel rushed because syllables are produced quickly, because pauses are short, or because too much information arrives without a useful boundary. These are related but different possibilities. Measuring only the total duration can hide the source of the difficulty.
Suppose a hypothetical recording contains 48 counted syllables over twelve seconds, including two seconds of pauses under an explicitly defined pause rule. Its overall speech rate is 48/12 = 4 syllables per second. If the articulation-time denominator excludes those two seconds, the corresponding rate is 48/10 = 4.8 syllables per second.
Keep the denominator in the label
Do not report both numbers as simply speed. One includes the defined pauses and the other excludes them. A comparison becomes misleading when one recording uses total time and another uses articulation time. State how syllables were counted, what counted as a pause and whether false starts were included.
Now consider a second twelve-second recording containing the same 48 syllables but four seconds of defined pauses. Its overall rate remains four syllables per second, while the articulation-time rate becomes 48/8 = 6. The identical overall rate conceals faster stretches of speech separated by longer gaps.
Neither pattern is automatically better. Long gaps may give a listener time to act or may interrupt a closely connected idea. Fast articulation may be intelligible to a familiar audience but difficult with unfamiliar names. Test the actual listening task instead of assigning a universal ideal number.
A useful repair is to add a boundary at a decision point while leaving a familiar phrase intact. For a spoken menu, separate options clearly and give enough time for selection. For a narrative, preserve the relationships that make the sequence coherent. “Slow down” becomes actionable only when it identifies which part should change and why.
Rhythm should likewise be investigated rather than reduced to a slogan about a language. Real speech varies with material, speaker and activity. Start with measured durations and a defined comparison, then ask what regularities the evidence supports.
CHAPTER 9 OF 14 · Check recordings, rates and listening
9. Design a listening comparison that can answer its question
A small experiment begins with a contrast. Suppose you want to test whether a particular placement of prominence helps listeners identify a corrected destination. Prepare comparable recordings with the same words and different intended prominence, then ask a question about destination rather than a general preference question.
Keep the surrounding context available. A correction is a response to an alternative; without that alternative, listeners may interpret the sentence differently. If the task supplies no prior misunderstanding, it may not test the communicative function you intended.
Work through a hypothetical result
Imagine ten listeners each hear four items in each of two conditions. That gives forty responses per condition. In condition A, thirty responses identify the intended correction; in condition B, twenty-two do. The descriptive success proportions are 75% and 55%, a difference of twenty percentage points.
The forty responses are not forty independent people. Each listener contributes several observations, and each item may have properties that affect difficulty. Those dependencies matter for any inferential analysis. The simple percentages describe this constructed dataset; they do not establish a population effect or a guaranteed teaching benefit.
Counterbalance order where possible. If everyone hears A after B, familiarity may contribute to better performance. Avoid putting the same item immediately beside its alternate version unless direct comparison is the intended task. Such pairing can reveal the experimental manipulation and change how people listen.
A manipulation check asks whether the recordings actually differ in the intended way. If you changed prominence but also slowed the entire sentence and raised playback volume, the outcome cannot be attributed cleanly to prominence alone. Record those differences and narrow the claim accordingly.
When a practical classroom comparison lacks experimental control, it can still guide revision. Describe it as a local usability check. The useful conclusion may be that three listeners repeatedly misunderstood one instruction, which gives you a concrete problem to repair without pretending to have discovered a universal law.
CHAPTER 10 OF 14 · Annotate, repair and test transfer
10. Annotate disagreements instead of hiding them
An annotation is a decision made under a definition. Two people may disagree about whether a word is prominent, where a phrase ends or which contour label best fits an utterance. The disagreement can reveal ambiguous evidence, insufficient training or a poorly specified category.
Begin with independent judgements on the same material. If observers discuss every item before recording a decision, the resulting agreement cannot tell you how reproducibly they apply the rules alone. Preserve the independent records before building a reconciled version.
Separate boundary and category disputes
For a fictional set of twenty items, two observers agree on the presence of a phrase boundary in sixteen. Their observed agreement is 16/20 = 80%. This number does not show whether they agree on the boundary’s location, whether one category dominates, or whether their shared interpretation is valid.
Inspect the four disagreements individually. Perhaps the recording ends too early to hear the phrase completion. Perhaps one observer treats any breath as a linguistic boundary. Those problems call for different repairs: more context in one case, clearer coding rules in the other.
The MIT ToBI course materials include uncertainty in labelling as an explicit topic. That is a useful research habit: uncertainty belongs in the record. A label should not become categorical merely because the software requires a field to be filled.
Maintain separate layers for words, timings, descriptive observations and theoretical labels when the task requires them. A beginner need not adopt a large annotation system immediately, but should define each mark clearly enough that someone else can understand the record.
After revising the codebook, check previously annotated examples that may be affected. Mixing old and new definitions without marking the change can create artificial differences within a dataset. Version the rules and explain which version generated the published results.
Agreement is evidence about a procedure. It is not a substitute for checking whether the procedure represents the phenomenon you intended to study.
CHAPTER 11 OF 14 · Annotate, repair and test transfer
11. Diagnose reading and presentation problems precisely
A reader can pronounce every word correctly yet make a passage difficult to follow. They may group words in an unhelpful way, emphasise an irrelevant contrast or rush through a change of idea. The repair begins with the listener’s difficulty rather than a broad demand to sound more expressive.
Use a short passage with a clear information structure. Ask a listener to identify the main point, the corrected detail or the sequence of actions. Compare their answer with the intended meaning. Then locate the stretch of speech where the relationship became unclear.
Change one feature and listen again
Suppose a fictional explanation says, “The result changed because the sample was warmer, not because the volume increased.” If the contrast is difficult to hear, first clarify the wording and grouping. Practise the two competing explanations as meaningful alternatives. Do not begin by telling the speaker to raise their voice throughout the sentence.
Record a second version and test it with someone who has not been coached to expect the answer. If the listener now identifies the intended cause, the revision has achieved something observable. If they still misunderstand, the problem may involve vocabulary or the explanation itself rather than delivery.
For longer presentations, mark changes of function: introducing a question, giving an example, explaining a calculation, stating a limitation and concluding an argument. These transitions create different listening demands. Prosody can support them, but clear verbal signposting remains important when an audience is unfamiliar with the topic.
Avoid using one accent as the hidden standard for competence. Decide whether you are assessing intelligibility, a particular communicative distinction or performance in a defined language variety. Tell the learner which criterion applies. A vague instruction to sound native can obscure the attainable skill that would improve communication.
Practise with new material after a familiar passage improves. A memorised performance may be fluent without demonstrating independent control. Ask the speaker to explain where they placed boundaries and why a particular detail deserved prominence. Their explanation helps distinguish purposeful choices from imitation.
CHAPTER 12 OF 14 · Annotate, repair and test transfer
12. Test transfer from speech to text and synthetic speech
A spoken correction can lose its emphasis when converted to text. A written instruction can acquire misleading phrasing when read by a synthetic voice. In both directions, preserve the intended relationship rather than assuming that identical words guarantee identical meaning.
Imagine a voice message: “Deliver the parcel on Thursday, not Tuesday.” A transcription that preserves the words still needs readable punctuation and an unambiguous date when the context requires one. A message saying only “Thursday” may be insufficient when several weeks are possible. Prosody cannot recover information that was never supplied.
Repair a synthetic instruction
Take the fictional text “Select option two after the tone.” Listen to the rendered audio. Does the voice separate the action from the timing condition in a way that preserves their connection? Can the intended listener repeat what to select and when? A valid markup file is not the same as a successful instruction.
The W3C SSML 1.1 specification defines controls that can influence properties such as pitch, rate and volume. These controls do not provide a universal mapping from a setting to a social meaning. Check the capabilities of the selected processor and evaluate its actual audio output.
The site’s spoken-language synthesis and translation guide develops this workflow across languages. It is especially useful when pronunciation, voice selection and wording must be checked together. The prosody root supplies the underlying distinction between an acoustic change and a communicative result.
Now change the context: remove the screen, introduce background noise, or give the instruction to a listener unfamiliar with the menu. Identify which information becomes fragile. Repair it through wording, pacing or an alternative format, then retest. Do not assume that success with headphones in a quiet room settles every intended use.
For an audio-only message, meaning must be available through sound. For a text alternative, important distinctions must be explicit in writing. Transfer succeeds when the necessary information survives the change of channel, even if the form changes substantially.
CHAPTER 13 OF 14 · Answer questions and continue learning
13. Answer common questions without overstating the evidence
Is prosody the same as pronunciation?
Pronunciation is a broader everyday term that can include individual speech sounds and prosodic patterns. For analysis, specify the level that matters. A problem identifying a consonant is different from a problem hearing which part of a sentence is being corrected, though both can affect understanding.
Does a higher pitch mean greater confidence?
Not by itself. Pitch depends on the speaker, the utterance, the linguistic pattern and the recording. A judgement about confidence requires evidence beyond one acoustic value. Describe what changed and test the relevant communicative outcome instead of treating a number as a personality measure.
Are pauses always mistakes?
No. Pauses can organise information, allow an action or form part of a normal speaking pattern. An interruption becomes a practical problem when it interferes with the task. Decide whether the issue is misplaced grouping, insufficient processing time or something else before prescribing a change.
Must every prominent word be louder?
No. Prominence can involve several cues and is interpreted relationally. Inspect duration, pitch and context where relevant. Increasing recording volume changes the signal level but does not necessarily create the intended linguistic contrast.
Can text alone prove how a sentence sounded?
No. Typography and punctuation can suggest an intended reading, but they do not preserve an actual performance. If the claim concerns what a speaker did, keep the recording or describe the evidential limitation. An invented reading should be labelled as an illustration.
Is an attractive pitch graph enough for a research report?
No. Include the question, recording conditions, analysis method, settings, selection rules and limitations. Explain what the graph supports and what it does not. A visual pattern becomes evidence only through a defensible connection to the claim.
How does gesture fit into the picture?
Speech occurs with visible activity in many settings. The Gesture Studies root explains how movement, timing and context contribute to meaning. When a recipient can see the speaker, test whether the interpretation depends on both channels rather than attributing every effect to the voice.
CHAPTER 14 OF 14 · Answer questions and continue learning
14. Build a learning route that ends in a checkable explanation
Begin with listening. Choose a short, appropriately obtained recording and one question about prominence or phrasing. Write the words, note the relevant context and describe the audible distinction in ordinary language. Resist collecting dozens of labels before you know what problem the analysis is meant to solve.
Next, make a comparison. Produce or locate a suitable alternative version, clearly distinguishing constructed examples from natural observations. Ask a listener a question about meaning. Preserve unexpected responses; they may expose an ambiguity that your own familiarity concealed.
Then add measurement where it answers a real question. Calculate a duration, a rate or a frequency relationship using the explicit methods above. Inspect the underlying recording and keep the settings with the result. Measurements should constrain the explanation, not decorate it.
Finally, change the context. Use a different sentence, another listener or an alternative delivery channel. Identify what must stay invariant: the corrected object, the action order, the intended grouping or the information being requested. Revise the message and test whether that relationship survives.
Choose the next source by the question
For formal transcription, consult the ToBI guidelines and the MIT teaching sequence, respecting their stated scope and material-use conditions. For acoustic analysis, use the Praat method documentation. For generated speech, begin with the SSML specification and test the actual processor.
A finished explanation should let another person locate the evidence, reproduce the calculation and understand the limits of the interpretation. Its practical value is clearer listening and clearer communication: knowing what a pattern contributes, recognising when that contribution is uncertain, and repairing a message when the context changes.