Skip to content

Quiz: Audio Signal Processing, Sound Classification, and Speech I/O

Test your understanding of audio signal processing, sound classification, and speech I/O with these review questions.

1. What is the Fast Fourier Transform (FFT)?

  1. A digital communication protocol for streaming digital audio between chips
  2. The baseline level of unwanted background sound in a recording
  3. A chip that converts a digital audio stream back into an analog voltage
  4. An algorithm that converts a block of time-based audio samples into a frequency spectrum
Show Answer

The correct answer is D. The Fast Fourier Transform is an algorithm that converts a block of time-based audio samples into a frequency spectrum, revealing which pitches are present in that block rather than just how loud the signal is. Option A describes the I2S audio interface, option B describes the noise floor, and option C describes a PCM5102 DAC.

Concept Tested: Fast Fourier Transform

2. How does sound event detection differ from sound classification?

  1. Sound classification assigns a single label to an entire clip of audio, while sound event detection identifies exactly when a specific sound starts and stops within a longer, continuous stream
  2. Sound classification identifies exactly when a sound starts and stops within a stream, while sound event detection assigns a single label to an entire clip
  3. Sound classification and sound event detection are two names for the same task
  4. Sound event detection only works with speech, while sound classification only works with environmental sounds
Show Answer

The correct answer is A. Sound classification is the general task of assigning a single label to a clip of audio from a fixed set of categories, while sound event detection goes further, identifying exactly when a specific sound starts and stops within a longer, continuous audio stream. Option B reverses these two roles, option C incorrectly treats them as identical, and option D invents a speech/environmental restriction the chapter does not describe.

Concept Tested: Sound Event Detection

3. How does wake word detection differ from a voice command?

  1. A voice command triggers listening for a wake word, reversing the usual order
  2. Wake word detection and voice commands are identical processes running on the same model
  3. A wake word is a lightweight model that listens constantly for one specific phrase, staying silent about everything else, while a voice command is a structured spoken instruction handled by a separate process that typically only runs after the wake word fires
  4. A wake word requires cloud processing, while a voice command always runs on-device
Show Answer

The correct answer is C. Wake word detection runs a lightweight model constantly, listening only for one particular phrase and staying silent about everything else, while a voice command is a structured spoken instruction typically handled by a separate, larger process that only runs after the wake word fires. Option A reverses the actual trigger order. Option B incorrectly treats them as identical, and option D invents a processing-location restriction the chapter does not make.

Concept Tested: Wake Word Detection

4. A project needs to accurately capture a tone with a maximum frequency of 6,000 Hz. Applying the sampling rule f_s ≥ 2·f_max described in this chapter, which sampling rate is the minimum adequate choice?

  1. 3,000 Hz
  2. 12,000 Hz
  3. 6,000 Hz
  4. 60,000 Hz
Show Answer

The correct answer is B. The Nyquist rule requires the sampling rate to be at least twice the maximum frequency being captured: 2 × 6,000 Hz = 12,000 Hz. Option A is below the tone frequency itself, option C exactly matches the tone frequency but not twice it, risking aliasing, and option D is far more than necessary, wasting data for no benefit at this tone frequency.

Concept Tested: Audio Sampling Rate

5. A wake-word model reports a confidence score of 0.63 for a given audio buffer, and the project's confidence threshold is set to 0.75. Based on this chapter, what should the program do?

  1. Treat this as a valid wake-word trigger and act immediately
  2. Automatically lower the threshold to 0.63 to match the observed score
  3. Switch from wake word detection to voice command recognition
  4. Not treat this as a trigger, since the score falls below the chosen threshold
Show Answer

The correct answer is D. A program typically only acts once a confidence score clears the chosen threshold; since 0.63 is below the 0.75 threshold, this buffer should not be treated as a valid trigger. Option A ignores the threshold entirely, option B invents an automatic adjustment mechanism the chapter does not describe, and option C confuses two entirely separate pipeline stages.

Concept Tested: Audio Confidence Score

6. A student is building a voice-controlled robot that must work reliably in a school gym with no internet access, and must not transmit any audio off the device for privacy reasons. Based on the chapter's comparison table, which processing approach should the student choose?

  1. On-device (edge) processing, since it works offline and audio never leaves the device
  2. Cloud processing, since it offers a larger and more accurate model
  3. A hybrid approach that always transmits audio to the cloud first, then falls back to on-device processing if the network fails
  4. Neither approach works without a wired Ethernet connection
Show Answer

The correct answer is A. The chapter's comparison table shows on-device (edge) processing working offline with audio that never leaves the device, exactly satisfying both the no-internet and privacy constraints described. Option B directly requires an internet connection and transmits audio off-device, violating both constraints. Option C still transmits audio off-device by design, and option D incorrectly claims neither approach can work without Ethernet.

Concept Tested: On Device Processing

7. A student records a 3,000 Hz tone using a microphone sampling at 4,000 Hz, and the resulting digitized audio sounds like a completely different, lower pitch than the original tone. What is the most likely explanation, based on this chapter?

  1. The microphone array's direction-sensing feature distorted the pitch
  2. The audio buffer size was set too large for the sampling rate
  3. The sampling rate is below the Nyquist minimum for this tone, causing aliasing that digitizes the sound as the wrong pitch
  4. The text-to-speech engine was accidentally enabled during recording
Show Answer

The correct answer is C. The Nyquist minimum for a 3,000 Hz tone is 6,000 Hz, but the microphone sampled at only 4,000 Hz — below that minimum — which the chapter identifies as causing aliasing, where higher-pitched sounds get digitized as the wrong pitch entirely. Option A misapplies an unrelated microphone-array feature, option B describes a timing issue not a pitch-distortion cause, and option D introduces an unrelated, irrelevant process.

Concept Tested: Audio Sampling Rate

8. A student wires a PCM5102 DAC and amplifier module to a Pi 5 and hears a faint buzz or hum through the speaker as soon as it powers on, even with no audio playing. Based on this chapter's guidance, what is the most likely cause and appropriate response?

  1. The DAC chip itself is almost certainly defective and should be replaced immediately
  2. The hum is a normal, unavoidable property of all I2S audio hardware and cannot be fixed
  3. The speaker output stage should be removed entirely and replaced with a piezo buzzer
  4. The hum is likely a grounding issue or a power supply shared with something noisy, and should be isolated one wire at a time using root cause analysis
Show Answer

The correct answer is D. The chapter's own encouragement note states that a faint buzz or hum the first time a DAC and amplifier are wired up is almost always a grounding issue or a shared noisy power supply, not a bad chip, and recommends applying root cause analysis to isolate one wire at a time. Option A jumps to the least likely explanation first. Option B incorrectly claims the hum is unfixable, and option C is an unnecessary overreaction that abandons the higher-quality audio chain for no reason.

Concept Tested: Amplifier Module

9. A teacher wants a Pi 5 project that can identify roughly which direction a sound came from in a noisy classroom with background chatter. Weighing the chapter's comparison of USB microphones and microphone arrays, which option should the teacher choose, and why?

  1. A USB microphone, because it is simpler to set up and direction sensing is not actually useful in a classroom
  2. A microphone array, because its two or more elements at known spacing let software compare timing differences to estimate sound direction, which a single USB microphone cannot do
  3. Either option works equally well, since direction sensing does not depend on the number of microphone elements
  4. A USB microphone, because it has a higher signal-to-noise ratio than any microphone array
Show Answer

The correct answer is B. The chapter's comparison table explicitly notes that direction sensing requires two or more microphone elements at known spacing, which only a microphone array provides — a single USB microphone cannot triangulate direction, since that capability requires at least two listening points. Option A dismisses a requirement the teacher explicitly needs. Option C contradicts the chapter's direct explanation of why direction sensing needs multiple elements, and option D makes an unsupported signal-to-noise claim.

Concept Tested: Microphone Array

10. A team wants to design a doorbell notification system that avoids false alerts from a delivery truck driving past (which sounds a bit like a knock) or a pedestrian walking by (who is not knocking). Which design best applies this chapter's concepts to solve that problem?

  1. Use multi-modal recognition, only triggering a notification when both the camera detects a person and the microphone's sound event detection detects a knock at roughly the same time
  2. Rely on sound event detection alone, triggering an alert any time a knock-like sound crosses the confidence threshold
  3. Rely on object detection alone, triggering an alert any time a person is visible in the camera frame
  4. Disable audio privacy considerations entirely so the system can record continuously as a workaround
Show Answer

The correct answer is A. Multi-modal recognition combines predictions from more than one sensor type to make a single, more reliable decision, and the chapter's own doorbell example describes exactly this scenario — combining "camera sees a person" with "microphone hears a knock" to cut down on false alerts that either sensor alone would produce. Option B is fooled by trucks that sound like knocks, option C is triggered by any passerby regardless of knocking, and option D abandons privacy responsibility without solving the false-alert problem at all.

Concept Tested: Multi Modal Recognition