Reader’s gateway. Two Secondary 1 students in Clementi record the same short spoken explanation for a Science presentation. Their phones rest beside each other on the same table. A classmate speaks at one ordinary volume, yet one recording produces a tall waveform and clear voice; the other looks quiet and distant. When they replay the files, the first student’s audio gets louder during a pause, including the scrape of a chair. The second file stays more even. Has the speaker’s voice changed, or did the devices change the sound after it entered the microphone?
The answer may be in the recording chain, especially automatic gain control. Phones and meeting applications can alter microphone sensitivity, amplify quieter input, reduce louder input, suppress background noise or compress dynamic range. A waveform shown in an editor describes a recorded signal, not a calibrated measurement of the sound pressure that existed in the room. Two devices can record different levels from one unchanged source. That matters for classroom audio investigations, oral presentations, listening exercises and any claim based on “the graph shows it was louder.”
The answer in about 50 seconds
A microphone converts sound-pressure changes into an electrical or digital signal. Before a file is saved, the device or application may apply amplification, automatic gain control, noise suppression, compression or limiting. Afterward, playback software may apply its own volume changes. These processes can make identical classroom sounds appear at different amplitudes even when the speaker has not changed. A taller waveform does not prove a louder real-world sound. To compare recordings, keep microphone type, distance, direction and processing as consistent as possible; inspect whether automatic mic sensitivity is enabled; and do not use an ordinary voice recorder as a calibrated decibel meter.
Learner: compare the recording settings before explaining the waveform. Parent: do not judge an oral presentation’s clarity solely from phone speaker volume. Teacher: distinguish speech intelligibility, recorded digital level and physical sound level before using audio as evidence. This is a useful case study in how a device’s attempt to improve communication can complicate measurement.
1. The recording is not the room
Sound begins as a physical event. A speaker’s vocal folds and mouth create pressure variations that travel through air and interact with walls, furniture, people and distance. A microphone responds to some of those variations according to its design, direction and position. The device converts that response into a signal. Electronics and software may then alter that signal before a pupil sees a waveform or hears playback through headphones.
These layers are easy to forget because a recording seems like a direct copy of the event. It is not. It is a representation produced by a chain of choices. Different microphones have different sensitivities and frequency responses. One may favour speech frequencies; another may capture more room ambience. A phone case can partly obstruct an opening. A hand resting beside the microphone may alter what it receives. A call application may treat background sounds differently from speech. The apparent loudness at playback also depends on speaker or headphone output and the listener’s volume setting.
A student comparing two files may therefore be comparing two chains rather than two performances. To answer the scientifically interesting question—”Which event was louder in the room?”—the chains must be made comparable or replaced with an appropriate calibrated measuring instrument. To answer a different question—”Which recording is easier for classmates to understand?”—audio processing may be beneficial and should be evaluated by intelligibility rather than physical sound level.
This distinction is a first-principles lesson about measurement: the number or image on a screen is an output of an instrument-plus-processing system. A strong investigation traces the path from real event to displayed value. A weak one assumes that because a graph is precise-looking, it is measuring the intended quantity directly.
2. Follow the signal from voice to file
The sequence is: voice and room → sound pressure at the microphone → microphone transducer → preamplifier and gain → optional automatic processing → analogue-to-digital conversion → stored audio → playback processing → listener. The order and presence of these stages vary by hardware and software; some processing occurs after digitisation rather than before it. The conceptual point is that the path contains opportunities for change between the original pressure wave and the displayed waveform.
The microphone has a sensitivity: a given sound pressure produces a corresponding electrical response under specified conditions. The preamplifier raises that response so the recording system can use it effectively. Gain can be fixed by a manual control or adjusted automatically. If gain is too low, quiet speech may be buried in background or electronic noise. If it is too high, loud peaks may overload part of the chain and distort. Neither problem can necessarily be repaired by turning up the volume during playback.
Once the signal is digital, software can scale it, filter it, compress its dynamic range, suppress selected sounds or encode it in a file. An audio editor might show the waveform after those transformations. If one device doubles the amplitude of its signal relative to another, the peaks may appear twice as tall even though both microphones received comparable sound pressure. A convenient representation has changed, not necessarily the physical event.
A useful class diagram puts the real-world event at the left and the final graph at the right. Pupils label which stages might differ between phones. This makes their hypothesis testable: if the phones are swapped but the difference follows the phone, device processing is plausible. If the difference follows position in the room, geometry and acoustics may be responsible. If both change, there may be more than one cause.
3. What automatic gain control actually tries to achieve
Automatic gain control, or AGC, adjusts amplification to keep a recording or communication signal within a useful level range. Its purpose is often practical. A person moves slightly away from the microphone, speaks more quietly or raises their voice, and software tries to make the voice remain audible without forcing listeners to adjust their speakers continuously. The control may react over time rather than instantly, because it must estimate a useful input level from changing sound.
Imagine a speaker begins with a normal sentence, pauses, whispers, then speaks forcefully. With fixed input gain, the corresponding waveform might have a moderate section, a much smaller section and a high-amplitude section. With some forms of AGC, the system may increase amplification during quieter speech and reduce it during the loud section. The saved waveform can therefore look more even. That is useful for conversation but less useful if the student intended to measure how loudly the speaker actually performed each part.
Timing creates a distinctive symptom. After a loud sound ends, an automatic controller may gradually raise gain while listening for the next voice. A chair scrape or air-conditioning hum during the pause can become surprisingly prominent. When speech resumes, the controller may reduce gain again. This can create audible “pumping” or changing noise. However, not every change during a pause is AGC. Noise suppression, microphone direction, movement and playback processing can produce other effects. The learner needs a comparison, not a diagnosis based on one clue.
The control strategy also varies across applications. A voice memo recorder, online meeting app and video camera app may use different defaults even on the same phone. Microsoft Teams, for example, documents an option to adjust microphone sensitivity automatically. That is evidence such a feature exists in a mainstream communications product; it is not proof that every phone recorder applies the same algorithm.
4. The worked example: twice the amplitude is not twice the acoustic pressure
Suppose an invented experiment plays a constant, moderate test tone through one classroom speaker while two phones record from comparable positions. Device A produces a peak sample magnitude of 0.20 relative to its digital full-scale limit; Device B, with greater recording gain, produces 0.40. Both use the same normalisation convention and neither clips. The digital peak of B is twice A. That difference alone tells us nothing about which location experienced greater acoustic sound pressure, because the electronics and processing are not calibrated to each other.
In digital audio, a doubling of amplitude corresponds to about 6.02 dB when expressed as an amplitude-level difference: 20 log10(2) ≈ 6.02. This is a comparison of signal amplitudes under a stated reference. It must not be confused with a direct measurement of room sound pressure level in dB SPL. The familiar “dB” unit appears in many settings, but every level needs a reference quantity and often a weighting or measurement procedure. A value in dBFS—the digital system’s level relative to full scale—is not automatically a value in dBA at the listener’s ear.
Now turn on automatic gain in one recording application, if the software allows that safely and clearly. A weaker incoming signal might be amplified more aggressively; a stronger one might be held back. The ratio of displayed peaks can change again without any change to the speaker or room. The graph shows the end of a processing chain, not an unmediated physical reading.
For a Secondary learner, the calculation is optional but revealing. It connects logarithms, ratios and measurement interpretation. The more important understanding is that multiplication by gain changes the recorded waveform. A learner should be able to explain how two files can differ mathematically while the real-world sound source remains the same.
5. Automatic gain is not the same as normalisation
Students often group all loudness adjustments into one category, but different operations behave differently. Normalisation typically applies a calculated gain change to an existing selection so that its highest peak reaches a chosen target. Audacity’s manual describes peak normalisation as setting the loudest point to a specified level, often leaving headroom below digital full scale. In a straightforward peak-normalise operation, one gain factor can be applied across the selected audio.
Dynamic-range compression changes the relationship between quiet and loud portions, especially above a threshold. Audacity documents compressor controls such as threshold and ratio. Compression can pull down loud parts relative to quiet parts, and make-up gain can raise the result. A recording may sound more consistently audible after compression without preserving the original amplitude differences. AGC is a broader category of level-control behaviour that often changes gain automatically during recording or communication. A particular product may implement it with processing related to compression, but the words do not name exactly identical operations in every application.
Noise suppression aims to reduce unwanted background sounds. It may use frequency filtering, statistical estimates or machine-learning methods. A suppressed fan or keyboard tap can make speech seem clearer, yet that is not the same operation as calibrating sound pressure. Limiting prevents or constrains high peaks in some systems, and clipping occurs when a stage is overloaded and cannot represent the peak faithfully. Turning the saved file down afterward does not restore information lost through clipping.
There is another term to untangle: audio file compression can also mean data compression, such as making a file smaller through an encoding format. That is not identical to dynamic-range compression. A small encoded file can contain wide dynamic range; a large uncompressed file can contain audio that has already been level-compressed. Pupils who separate these mechanisms are better equipped to interpret audio software menus and to explain what an editing step did to their evidence.
6. Distance, direction and the classroom itself still matter
Even with processing disabled, two microphones beside each other may not capture identical signals. Their openings could point in different directions. One phone might be slightly shielded by a book. A speaker may face one device more directly. Reflections from a wall, tabletop or window can strengthen or cancel some frequencies at particular positions. Background noise can arrive from a different direction than the voice. These are physical changes to the pressure field at each microphone, not software changes.
Distance matters as well. In an ideal free field with a compact source, sound intensity decreases approximately with the square of distance. Real classrooms are not free fields: reflections, background noise and the directional character of speech complicate the relationship. A pupil should not use a simple inverse-square calculation to predict exact recorded waveform peaks across desks without controlling the other factors. The formula can explain a tendency, not replace actual acoustic measurement in a reverberant room.
A small demonstration can separate causes. Place one phone at a fixed point; record a short consented speaker sample with the phone facing forward, then rotate it while keeping distance roughly constant. Next move it farther away. Keep the application and settings constant. The changes reveal that geometry affects recording. Only afterward compare different applications at the original position to investigate processing. Changing distance, direction and application simultaneously makes the outcome uninterpretable.
For learners, the broad insight is that the room, the microphone and the software all participate. A phone is not just a passive collector of “what happened.” It receives a local version of the acoustic field and transforms it. A trustworthy comparison needs a clear target: sound pressure, intelligibility, waveform shape or quality of the presentation. Each target may require a different method.
7. Why a quiet pause can become noisy in a recording
Return to the Clementi opening. The classmate speaks, pauses, and a chair scrape suddenly seems loud in one recording. A plausible explanation is an automatic controller recovering gain during the quieter interval. The control may raise amplification in an attempt to capture soft speech. If background noise is present, it is amplified too. Once the speaker resumes, the controller may lower gain to keep the voice within a target range.
But the same symptom can have other causes. The chair may have moved closer to one microphone. A noise-reduction algorithm may have stopped suppressing a transient sound. The phone might have switched microphones or altered its direction. One recording app might have normalised the whole file after recording. A good investigation distinguishes these hypotheses with controlled tests rather than assuming AGC whenever a background sound is prominent.
One diagnostic is to compare a longer quiet interval under otherwise stable conditions. If the background noise gradually rises without a corresponding physical increase, changing gain is plausible. Another is to use fixed-gain recording where available and safe, then compare. Yet ordinary smartphones may not expose the internal processing path clearly. If that control is unavailable, the honest conclusion is “the recorder may have applied automatic processing” rather than “we proved which algorithm changed the sound.”
This is also a lesson in causal inference for school Science. A visible or audible difference can be real while its explanation remains uncertain. The researcher should first establish a repeatable observation and then isolate factors. A beautiful waveform is not a substitute for a controlled comparison, and the best report distinguishes observations from proposed causes.
8. Why waveforms are useful but not calibrated sound meters
An audio waveform displays the variation of a recorded digital signal over time. It is useful for locating speech, pauses, sharp transients, distortion and differences in recorded level. It does not inherently express physical sound pressure in pascals or a weighted environmental sound level. Its vertical scale may be an arbitrary sample magnitude, a fraction of digital full scale or a logarithmic presentation chosen by the application.
A sound level meter has a different job. Proper instruments measure acoustic quantities using specified microphone characteristics, calibration, frequency weighting and time weighting. The U.S. National Institute for Occupational Safety and Health describes its Sound Level Meter app as a tool developed and tested for sound measurement, with important conditions and limits, including device and microphone considerations. That resource also makes clear that a validated sound measurement should not be inferred from an arbitrary consumer recording app.
The distinction matters in a school experiment asking whether two learning spaces have different noise exposure. Playing two recordings and deciding which “sounds louder” on one phone is not equivalent to measuring comparable A-weighted sound levels. The volume slider, playback hardware and post-processing alter the listener’s experience. Students can still investigate subjective intelligibility or recorded level, but they should name that outcome honestly instead of publishing unsupported decibel figures.
Likewise, digital full-scale units are relative to a file or recording system’s maximum representable level. A peak near zero dBFS indicates that the recorded signal approaches its digital ceiling; it does not say the classroom was close to a dangerous acoustic sound level. Conversely, a quiet-looking waveform could have been recorded in a physically loud room with very low gain. The word “level” needs a defined reference.
9. A safe, consent-based experiment using two devices
Begin with the research question: “Does changing the recording application affect the waveform level when the speaker and position are approximately unchanged?” This is a comparison of recording outputs, not a certified measurement of room loudness. Obtain teacher permission and the speaker’s consent. Use a prepared sentence, a quiet space and ordinary conversational volume. Do not record classmates who have not agreed or capture unrelated private conversation. Avoid loud test tones or potentially harmful sound exposure.
Place one device in a marked position and record the same sentence twice, once in each application if practical. Note the microphone choice, distance, direction, application version and any visible automatic-level or noise-suppression settings. Play the recordings back at the same listener volume, and inspect their waveforms in the same editor without applying automatic normalisation on import. If the apps use different codecs, note that too; this is a classroom comparison, not laboratory-grade isolation.
Next record a second trial with a visible automatic-level option changed, if the device permits it without compromising privacy or safety. Keep the sentence, distance and position the same. Do not expect perfect reproduction: people cannot speak identically each time. Repetition can help show whether the direction of difference is consistent. Record which settings you can verify and which internal processing steps remain unknown.
The report should separate three layers: observation (one waveform has higher peaks), controlled conditions (same device position and phrase, but possibly different processing), and interpretation (software gain is a plausible contributor). A stronger follow-up uses a stable loudspeaker signal at a safe level, if available and teacher-supervised, because that reduces variation in human delivery. The point is to design a comparison that narrows uncertainty without pretending an ordinary classroom is an acoustics laboratory.
10. The difference between a fair oral assessment and a fair audio measurement
For an English oral presentation, the goal may be to assess explanation, organisation, pronunciation and intelligibility. A recording with automatic gain can help classmates hear quieter speech. That is not inherently cheating or scientifically wrong; processing may be a legitimate accommodation or practical aid. The relevant issue is fairness and consistency of assessment conditions, especially if teachers are judging vocal projection using recordings rather than hearing the speaker in person.
For a Science investigation into sound amplitude, the goal is different. If the recorder changes gain independently, the student loses a stable relationship between original signal and output. That is why a feature beneficial for communication can undermine a particular measurement. The software is not malfunctioning; it is optimising for a different task. The question “Does this work?” must be followed by “Work for what?”
Teachers should tell pupils which outcome is being assessed. If the task is oral communication, learners may be allowed to record with standard voice-enhancement settings, but the criteria should reflect the final product and declared accommodations. If the task is acoustic comparison, document input gain and processing more carefully, or use suitable measuring equipment. If the task is media literacy, compare the results and ask why the representations differ.
This also matters for remote learning. A pupil with a poor microphone or noisy home environment may be less intelligible through a recording despite strong subject knowledge. Sound-processing variation is one reason to avoid turning recording quality into an unexamined proxy for understanding. Reasonable alternative submissions or teacher feedback can focus on the intended learning outcome rather than accidental device differences.
11. What the software settings can and cannot tell you
An option labelled “automatically adjust microphone sensitivity” suggests the application may control input level dynamically. Microsoft Teams documents such a setting in its device controls. An option called “noise suppression” points toward removing background sound. An option called “echo cancellation” addresses sound from speakers feeding back into the microphone. These names are useful clues, but they do not reveal every implementation detail inside a particular device or version.
Some phones apply hardware-level or operating-system processing before the application receives samples. A user may turn off one visible setting while another part of the chain continues to adapt the signal. An app may also change behaviour depending on whether it detects speech, music or a call. Therefore, “AGC off” in one menu should not be presented as proof that the entire path is fixed and scientifically calibrated. A measurement claim still needs appropriate calibration and documentation.
A student comparing settings should write what was actually observed: “In Application A, the automatic sensitivity option was enabled; in Application B, the control was not shown.” Avoid writing “Application B has no automatic gain” merely because its interface offers no checkbox. Absence of a visible switch is not evidence of absence of processing.
This is a useful digital habit beyond audio. The user interface exposes part of a system, not every operation under the hood. A well-designed classroom investigation distinguishes adjustable settings from hidden defaults, and recorded outputs from properties of the physical world. Students can work meaningfully even when the complete implementation is proprietary, provided they state what remains unknown.
12. Why higher waveform peaks do not guarantee better speech clarity
Loudness, intelligibility and quality overlap but are not identical. A strongly amplified recording may contain clipping, reverberation or background noise that makes words harder to understand. A quieter recording with a clean signal may be easier to transcribe after playback level is adjusted. Noise suppression may improve a voice in one room while removing consonant detail or musical sounds in another. The effect depends on the task and processing settings.
A clear explanation from a student includes pace, pronunciation, relevant vocabulary and structure. Increasing digital gain cannot repair a missing logical step. It can only change the recorded representation of the sound. Conversely, an excellent explanation can become difficult to follow when a recording system obscures quiet syllables or introduces pumping. This is why audio feedback should identify whether the barrier lies in the learner’s communication, the acoustic environment or the recording chain.
A practical listening check asks two or three consenting classmates to transcribe one short sentence from each file at a comfortable, matched playback level. Compare the missing words rather than voting only on which version sounds louder. That is an informal classroom activity, not a validated speech-intelligibility test. It can still demonstrate that a taller waveform is not a sufficient definition of a useful recording.
For pupils with hearing differences or language-processing needs, provide captions or a written transcript when appropriate. A clean sound file is helpful, but accessibility should not depend on perfect hearing or one device’s playback. The aim is to preserve the content of the lesson and offer more than one reliable route to it.
13. When sound pressure measurement is genuinely needed
Suppose pupils want to compare noise during silent reading and group discussion. They should define a measurement procedure that captures a relevant acoustic quantity. A validated sound level meter can report metrics such as time-averaged A-weighted levels, subject to its specifications and calibration. A recording can document the character of sound, with consent, but should not be treated as a substitute for instrumented measurement solely because it has a waveform.
Placement must be consistent and safe. A meter held near a noisy fan and another by a quiet window do not observe the same acoustic conditions. Duration matters: a short peak tells a different story from an average over a lesson. Weighting matters because human hearing responds differently across frequencies. The teacher should explain why a single “dB” number without metric and context is incomplete.
NIOSH’s materials distinguish its tested app and instrument conditions from general-purpose recordings. They also place sound measurement within a hearing-conservation context. A school should not use improvised high-volume experiments or expose pupils to risky noise simply to produce clearer differences. Use normal environmental levels, obtain appropriate supervision and follow the institution’s safety procedures.
If validated equipment is unavailable, revise the research question. Instead of claiming “Room A is five decibels louder than Room B,” investigate “How does the recorded waveform differ under these two documented conditions?” or “How accurately can classmates understand a sentence recorded at two distances?” Honest redesign preserves the educational value without pretending to measure what the equipment cannot support.
14. Common failure modes and quick repairs
Failure: the taller waveform is called a louder original event. Repair by identifying the waveform’s digital reference and checking gain and processing. A file-level peak cannot directly establish acoustic pressure.
Failure: two phones are compared without controlling geometry. Repair by using one marked position, consistent microphone direction and repeated trials, or acknowledge the uncontrolled variables.
Failure: automatic gain is mistaken for background-noise removal. Repair by separating level control from noise suppression; both can affect the result, but they have different purposes.
Failure: a student normalises two files before comparing original peak levels. Repair by keeping untouched originals and documenting each processing step. Normalisation can remove the amplitude difference under investigation.
Failure: a clipped recording is simply turned down. Repair by rerecording with appropriate input gain when possible. Reducing playback volume does not reconstruct flattened peaks.
Failure: the class records private conversations. Repair by using a scripted, consented voice or a safe non-speech signal, following school recording rules and deleting unneeded files.
Failure: an ordinary recorder is used to make precise dBA claims. Repair by choosing a calibrated sound measurement method or limiting the claim to recorded digital levels and listening outcomes.
Failure: a quiet pupil is graded as lacking knowledge because the recording is faint. Repair by separating recording conditions from content assessment and offering reasonable accessible alternatives.
15. A learner route: test one cause, not everything
Begin by writing the exact question and naming the output variable. If the aim is to understand AGC, a useful question is whether recorded level changes during a stable, consented audio source when one visible automatic-sensitivity setting changes. The variable is not “loudness of the classroom” but a defined measure in the file, such as peak magnitude or average signal level. Avoid combining physical sound-level measurement and recording quality into one vague score.
Prepare a small table with columns for application, microphone, automatic setting, distance, orientation, repeated-trial number and observed waveform feature. Keep the original files. Mark sections that contain speech, silence and transients. Compare not just one dramatic peak but the pattern over time. If noise rises during a pause, note that it is consistent with some gain-changing behaviour but not a unique fingerprint of AGC.
Next explain the pathway in a diagram. A sentence might say: “The speaker produced sound waves, which the microphone converted to a signal. The application may have changed the signal’s gain, so the saved amplitudes cannot alone establish different sound pressure levels.” This is the conceptual target. Once the learner can say it accurately, introduce optional calculations in dBFS or dynamic range.
Finally write a limitation. “We used different applications and could not verify all internal processing settings” is useful. So is “Human delivery was not identical across trials.” Strong Science writing is not a list of excuses; it is an account of what the test can and cannot establish. The next experimental step should answer the largest unresolved question rather than merely producing more audio files.
16. A parent route: listen to the idea before judging the recording
Parents reviewing a child’s recorded oral presentation may notice uneven volume and assume the child failed to speak clearly. Sometimes that is true; sometimes the recording placement or automatic processing made an otherwise understandable presentation sound inconsistent. Ask the child to explain the same short section directly in a quiet room. If the content is clear in person but difficult to hear on the file, investigate the recording chain before deciding the learning itself is weak.
Practical improvements include choosing a quiet recording area, keeping the microphone unobstructed and at a consistent distance, testing one short segment, and checking playback with the actual submission device. Avoid repeatedly increasing gain until sound distorts. If a communications app offers automatic sensitivity or noise control, a small test can show whether it helps speech intelligibility. The best setting for a lesson call may differ from the best setting for a music performance or a controlled Science investigation.
Parents should also respect consent and privacy. A home practice recording for a child’s own use differs from distributing classroom voices publicly. If school policy controls recording, follow it. For younger children, a spoken explanation and feedback from an adult can be more educational than an elaborate editing session. The recording is a tool to preserve and review communication, not a compulsory performance arena.
A helpful prompt is “Which part was hard to understand: your explanation or the sound quality?” That question opens a diagnosis rather than a judgement. If the child can identify a missing explanation, revise the words. If the problem is clipped or fluctuating audio, revise the recording method. This separation prevents a technical limitation from turning into misplaced doubt about the pupil’s ability.
17. A teacher route: teach the signal chain explicitly
Draw the pathway from source to listener on the board and ask students to place possible changes at each stage. Place “speaker turns head” before the microphone, “automatic input gain” in recording processing, “normalise the file” after capture and “listener moves volume slider” at playback. Then show how one physical event can become several digital outputs. The ordering makes otherwise confusing vocabulary useful.
A Science assessment can ask students to interpret controlled and uncontrolled conditions. An English assessment can ask how recording quality changes audience access while evaluating the actual organisation of speech separately. A Mathematics lesson can explore ratios and decibel calculations without claiming a digital peak measures environmental safety. A Computing lesson can consider audio pipelines and software settings. The mechanism becomes a bridge rather than a disconnected technical curiosity.
Model responsible evidence language. A pupil who writes “Phone B was louder” should be asked whether that refers to the original sound, the file or playback. An improved answer might be “Phone B’s saved waveform had greater peak amplitude in the editor, but the microphones and gain settings differed, so we cannot infer a higher sound pressure level.” That is an excellent scientific conclusion even if it feels less exciting than the first.
For class recording tasks, publish clear permissions and accessible submission routes. Provide a quiet option where feasible, and avoid evaluating presentation quality through inconsistent device defaults alone. If the task genuinely requires acoustic data, specify suitable measurement equipment and safety arrangements. The teacher’s responsibility is to make the instrument fit the educational question.
18. Evidence, caveats and the wider meaning
Microsoft’s device documentation confirms that Teams can automatically adjust microphone sensitivity, demonstrating a real-world use of adaptive input level. Audacity’s official manuals explain peak normalisation and dynamic-range compression as distinct operations on recorded audio. NIOSH’s sound-level materials explain the role of calibration and specified equipment for acoustic measurement. Together these sources support the central mechanism: recorded or played-back amplitude is partly a function of the system that captured and processed sound.
These sources do not prove what an unspecified phone did in the invented Clementi classroom. We did not inspect the devices or perform a controlled experiment. Nor should a classroom activity be presented as a validated test of every audio-processing algorithm. The lesson is to check the chain, identify the relevant variable and describe uncertainty when processing steps cannot be observed.
This matters beyond sound. A camera applies exposure and processing before showing an image. A spreadsheet bins values before displaying a histogram. A search result presents a condensed representation of a longer source. In each case a useful output is also a transformation. Trust grows when learners can follow the transformation well enough to know what it preserves and what it changes.
The confident takeaway is therefore not “phones lie.” Phones do many tasks well, and adaptive recording can make speech more usable. The takeaway is that a device optimised for communication is not automatically a calibrated measuring instrument. When the task changes, the learner must check whether the tool’s hidden improvements have changed the evidence they intended to collect.
Frequently asked questions
Why do two recordings of the same voice look different in an audio editor?
Different microphones, placement, preamplifier gain, automatic processing, file conversion or editing settings may change the digital waveform. The difference does not by itself establish a change in the original speaker’s sound level.
What is automatic gain control?
It is a system that adjusts amplification to keep an input signal within a useful range as the input changes. Specific algorithms and controls vary across products.
Is AGC always bad for school recordings?
No. It can improve ordinary speech communication. It becomes problematic when a student assumes the processed waveform faithfully preserves original amplitude differences for measurement.
Is normalisation the same as AGC?
Not necessarily. Peak normalisation usually applies a calculated gain change to a selected recording, while AGC can change gain dynamically during capture or communication. Compression and noise suppression are additional operations.
Can I measure classroom decibels from a voice memo waveform?
Not reliably without appropriate calibration and measurement procedures. A voice memo is not automatically an acoustic sound level meter, and dBFS is not dBA.
Why does background noise get louder when the speaker stops?
Some automatic level-control systems may raise gain during quiet passages, making background noise more prominent. Other causes are possible, so test rather than diagnosing from that symptom alone.
If I turn AGC off, are my results fully accurate?
No. Microphone sensitivity, distance, direction, room acoustics, other processing and calibration still affect the recording. Fixed gain is useful for some comparisons but is not equivalent to a certified sound measurement.
How can teachers assess oral work fairly when audio quality differs?
Separate the intended communication skill from accidental device-quality problems, provide clear recording guidance and reasonable alternative or accessible submission routes, and specify technical requirements only when they are part of the learning objective.
Sources and further reading
- Microsoft Support: Manage Your Device Settings in Microsoft Teams. Documents automatic microphone sensitivity and noise-control settings in an everyday communications tool.
- Audacity Manual: Normalize. Peak adjustment and its limits.
- Audacity Manual: Compressor. Dynamic-range control and ratio/threshold behaviour.
- NIOSH Science Bulletin: New NIOSH Sound Level Meter App. Measurement, calibration conditions and the distinction between validated sound-level tools and general audio apps.
- Australian Education Research Organisation: Teaching for How Students Learn. General teaching rationale for making hidden mechanisms explicit and testing understanding.
Continue through eduKateSG
Explore the How X Works hub and How Education Works for wider learning routes. For the related but distinct problem of how classroom distance and reverberation affect listeners, read Why the Teacher Sounds Clear at the Front but Not at the Back. That article owns the room-to-listener acoustic pathway; this one owns the microphone-to-file gain and processing pathway.
Editorial boundary: the Clementi recordings and numerical digital peaks are teaching examples, not measurements of real pupils, equipment or classrooms. Do not use ordinary audio recordings as a substitute for hearing-safety or compliance measurements.