📚 Year 12 CAIE Computer Science: Speaking and Listening Tech Revision | 口语/听力备考专项
In the world of computing, a machine's ability to 'speak' and 'listen' is not about natural intelligence but about capturing, processing and producing audio signals. For your Year 12 CAIE Computer Science exam, this topic bridges data representation, hardware and communication systems. Understanding how computers handle sound – from a simple microphone input to synthesised speech – will help you answer both theoretical and calculation-based questions with confidence. This article unpacks the core concepts you need, presenting paired explanations in English and Chinese to reinforce key terminology and processes.
在计算机的世界里，机器的'说话'与'听力'并非指自然智能，而是关于捕获、处理和生成音频信号。对于你的 Year 12 CAIE 计算机科学考试而言，这一主题将数据表示、硬件和通信系统串联起来。理解计算机如何处理声音——从简单的麦克风输入到合成的语音——将帮助你自信地回答理论与计算类题目。本文梳理了你需要掌握的核心概念，以中英配对的方式强化关键术语与过程。
1. The Computer’s 'Ears': Sound Input Hardware | 计算机的'耳朵'：声音输入硬件
To 'listen', a computer needs a microphone, which converts sound waves into analogue electrical signals. Because the CPU only processes digital data, an Analogue-to-Digital Converter (ADC) is essential. The ADC samples the analogue signal at regular intervals and quantises each sample into a discrete digital value. Key parameters here are sampling rate (measured in Hz) and sampling resolution or bit depth.
要'听'，计算机需要麦克风，它将声波转换为模拟电信号。由于 CPU 只处理数字数据，模数转换器（ADC）必不可少。ADC 以固定间隔对模拟信号采样，并将每个样本量化为离散的数字值。这里的关键参数是采样率（以 Hz 为单位）和采样分辨率或称位深度。
Most modern interfaces use an integrated sound card that contains the ADC, controlling the conversion process and transferring digital audio data to main memory via Direct Memory Access (DMA). For CAIE, you should be able to describe the role of the microphone, ADC and sound card in sound capture.
大多数现代接口使用集成的声卡，它包含 ADC，控制转换过程并通过直接存储器访问（DMA）将数字音频数据传输到主存。对于 CAIE，你需要能够描述麦克风、ADC 和声卡在声音捕获中的作用。
The quality of the 'hearing' primarily depends on the sampling rate and bit depth. A higher sampling rate captures higher frequencies, while a larger bit depth reduces quantisation noise, making the digital recording closer to the original analogue wave.
'听力'的质量主要取决于采样率和位深度。更高的采样率能捕获更高频的声音，而更大的位深度可减小量化噪声，使数字录音更接近原始模拟波形。
2. The Computer’s 'Voice': Sound Output Hardware | 计算机的'声音'：声音输出硬件
When a computer 'speaks', it reverses the input process. Digital audio data is read from storage or memory and sent to a Digital-to-Analogue Converter (DAC). The DAC reconstructs a staircase-like analogue signal from the discrete digital values. A low-pass filter then smooths this signal before it is amplified and sent to a speaker or headphones.
当计算机'说话'时，它会反转输入过程。数字音频数据从存储器或内存中读取，并发送至数模转换器（DAC）。DAC 将离散的数字值重建为阶梯状的模拟信号，再由低通滤波器平滑该信号，最后经放大送至扬声器或耳机。
Again, the sound card plays a central role, hosting the DAC and often offering multiple channels for surround sound. In exam questions, you may be asked to compare the data rates, explain the need for buffering or identify the sequence of components that leads from a stored .wav file to audible sound.
声卡再次扮演核心角色，它承载 DAC 并通常提供多声道以实现环绕声。在试题中，你可能需要比较数据速率、解释缓冲的需求，或识别从存储的 .wav 文件到可听声音之间所涉及的组件序列。
For output, the sampling rate and resolution used during recording determine the playback quality. If playback uses a lower sampling rate, the sound will be muffled; if the bit depth is too low, listeners may notice background hiss due to quantisation error.
对于输出，录制时使用的采样率和分辨率决定了回放质量。如果以较低采样率回放，声音会发闷；如果位深度过低，听众可能会因量化误差而察觉到背景咝咝声。
3. Digital Representation of Sound | 声音的数字表示
Sound in a computer is stored as a sequence of binary numbers, each representing the amplitude of the sound wave at a specific instant. The two fundamental parameters are sampling frequency (fₛ) and bit depth (n). The sampling frequency must be at least twice the highest frequency in the signal to satisfy the Nyquist theorem, otherwise aliasing occurs.
在计算机中，声音以一系列二进制数的形式存储，每个数代表声波在特定瞬间的振幅。两个基本参数是采样频率（fₛ）和位深度（n）。根据奈奎斯特定理，采样频率须至少为信号中最高频率的两倍，否则会出现混叠现象。
Sampling rate ≥ 2 × Highest Frequency (Nyquist Criterion) | 采样率 ≥ 2 × 最高频率 (奈奎斯特准则)
Standard CD quality uses a sampling rate of 44.1 kHz and 16-bit resolution. A quick calculation for file size of uncompressed mono audio is:
标准 CD 音质使用 44.1 kHz 的采样率和 16 位分辨率。计算未压缩单声道音频文件大小的方法如下：
File size (bits) = sampling rate (Hz) × bit depth × duration (seconds) | 文件大小(位) = 采样率(Hz) × 位深度 × 时长(秒)
Always remember to convert bits to bytes by dividing by 8, and further to kilobytes or megabytes as required. For stereo files, multiply by 2 for two channels.
务必记住将位除以 8 转换为字节，再根据需要转换为千字节或兆字节。立体声文件则需乘以 2 以表示两个声道。
| Parameter (参数) | Symbol (符号) | Effect on Quality & Size (对质量与大小的影响) |
|---|---|---|
| Sampling rate | f | Higher → wider frequency range, larger file size |
| Bit depth | n | Higher → less quantisation noise, larger file size |
| Duration | t | Longer → proportionally larger size |
| Channels | ch | Mono=1, Stereo=2; stereo doubles file size |
4. Speech Recognition: How Computers 'Understand' Spoken Words | 语音识别：计算机如何'听懂'口语
Speech recognition is the process of converting spoken language into text or commands. It involves several stages: acoustic analysis, feature extraction, pattern matching against acoustic and language models, and output generation. The microphone captures sound, which is then digitised and broken into short frames. Features like Mel-Frequency Cepstral Coefficients (MFCCs) are extracted to represent the spectral shape.
语音识别是将口语转换为文字或指令的过程。它涉及多个阶段：声学分析、特征提取、基于声学模型和语言模型的模式匹配，以及输出生成。麦克风捕获声音后，将其数字化并分割为短帧，提取诸如梅尔频率倒谱系数（MFCC）等特征来表示频谱形状。
The system uses a Hidden Markov Model (HMM) or deep neural network to map these features to phonemes, the smallest units of sound that distinguish words. Language models then predict the most probable word sequences. This allows your smartphone to transcribe a voice message or a virtual assistant to execute a spoken command.
系统使用隐马尔可夫模型（HMM）或深度神经网络将这些特征映射到音素，即区分单词的最小声音单元。然后语言模型预测最可能的词序列，使你的智能手机能够转写语音消息，或让虚拟助手执行语音指令。
CAIE exam questions on speech recognition often focus on the need for training (speaker-dependent vs. speaker-independent systems), the impact of background noise, and the reasons why speech recognition remains computationally intensive. You should also be aware of real-time applications and the difference between speech recognition and voice recognition (the latter identifies the speaker, not the words).
CAIE 关于语音识别的试题常关注训练的需求（特定人 vs. 非特定人系统）、背景噪声的影响，以及语音识别为何仍然计算密集型。你还应意识到实时应用，以及语音识别与声纹识别（后者识别说话人而非话语）之间的区别。
5. Speech Synthesis: Teaching Computers to 'Speak' | 语音合成：教计算机'说话'
Speech synthesis, often called text-to-speech (TTS), is the artificial production of human speech. Early systems used concatenative synthesis, which joined pre-recorded fragments of speech (phones, diphones). Modern TTS relies on parametric synthesis using Hidden Markov Models or deep learning, generating waveforms directly from linguistic features.
语音合成，常称为文语转换（TTS），是人工生成人类语音。早期系统使用拼接合成，将预先录制的语音片段（音素、双音子）拼接起来。现代 TTS 依靠基于隐马尔可夫模型或深度学习的参数合成，直接从语言学特征生成波形。
The process typically starts with text normalisation (converting abbreviations, numbers into words), followed by linguistic analysis to determine pronunciation and prosody (pitch, duration, stress). Then, a waveform generator produces the audio. Naturalness and intelligibility are key evaluation metrics.
该过程通常从文本规范化开始（将缩写、数字转换为完整单词），随后进行语言学分析以确定发音和韵律（音高、时长、重音）。然后，波形生成器产生音频。自然度和可懂度是关键的评价指标。
In the context of CAIE, you should be ready to describe how a screen reader uses TTS to assist visually impaired users, and to discuss the advantages and limitations of different synthesis methods. The concept of a phonetic dictionary and the role of a DSP (Digital Signal Processor) in real-time synthesis are also examinable.
在 CAIE 语境中，你应能描述屏幕阅读器如何使用 TTS 辅助视障用户，并讨论不同合成方法的优缺点。发音词典的概念、DSP（数字信号处理器）在实时合成中的作用也在考查范围内。
6. The Sound Card and Digital Signal Processing | 声卡与数字信号处理
The sound card is the dedicated hardware that handles all audio input and output. It houses the ADC, DAC, a dedicated DSP chip, and various input/output ports. The DSP can compress, decompress and apply effects (like echo cancellation, equalisation) in real time, offloading these tasks from the CPU. This is vital for VoIP calls, gaming and professional audio production.
声卡是负责所有音频输入输出的专用硬件。它包含 ADC、DAC、专用 DSP 芯片和各种输入/输出端口。DSP 能够实时压缩、解压缩并应用效果（如回声消除、均衡），将这些任务从 CPU 卸载。这对 VoIP 通话、游戏和专业音频制作至关重要。
In simpler systems, the sound functions may be integrated into the motherboard chipset, sharing memory and CPU resources. You should understand the trade-off: integrated audio is cheaper and consumes less power, but may offer lower quality and higher latency compared to a discrete sound card.
在较简单的系统中，音频功能可能集成在主板芯片组中，共享内存和 CPU 资源。你应理解其权衡：集成音频成本更低、功耗更小，但与独立声卡相比，可能质量较低且延迟较高。
Exam questions may ask you to identify which component performs the conversion (ADC for mic, DAC for speaker) and to explain why a sound card is needed even when the CPU is powerful. The key reason: real-time audio requires dedicated, predictable timing that a general-purpose CPU cannot always guarantee.
考题可能要求你识别哪个组件执行转换（麦克风用 ADC，扬声器用 DAC），并解释即使 CPU 强大为何仍需声卡。关键原因：实时音频需要专用的、可预测的时序，这是通用 CPU 未必总能保证的。
7. Audio File Formats and Compression | 音频文件格式与压缩
Uncompressed formats like .wav or .aiff store raw PCM (Pulse Code Modulation) data, which are large but lossless. Compressed formats can be lossless (e.g., FLAC) or lossy (e.g., MP3, AAC). Lossy compression exploits psychoacoustic principles, discarding sounds that the human ear is less sensitive to, such as frequencies masked by louder sounds.
未压缩格式如 .wav 或 .aiff 存储原始 PCM（脉冲编码调制）数据，体积大但无损。压缩格式可分为无损（如 FLAC）和有损（如 MP3、AAC）。有损压缩利用心理声学原理，丢弃人耳不太敏感的声音，例如被较响声音掩蔽的频率。
When comparing file formats, a classic exam task is to calculate the compression ratio or to explain why a 5-minute .wav file might be 50 MB while the same recording as an MP3 is only 5 MB. You should also be aware of the trade-off: lossy compression reduces quality, introducing artefacts at very low bitrates.
比较文件格式时，典型的考题任务是计算压缩比，或解释为何一个 5 分钟的 .wav 文件有 50 MB，而同一录音的 MP3 文件仅 5 MB。你还应意识到折中：有损压缩会降低质量，在极低比特率下会引入失真。
MIDI (Musical Instrument Digital Interface) is a completely different representation: it stores instructions (like notes, velocity, instrument) rather than waveforms. MIDI files are extremely small but rely on the playback device's synthesiser, resulting in inconsistent sound across devices.
MIDI（乐器数字接口）是完全不同的表示形式：它存储指令（如音符、力度、乐器）而非波形。MIDI 文件非常小，但依赖播放设备的合成器，导致不同设备上声音不一致。
8. Applications and Ethical Considerations | 应用与伦理考量
Speaking and listening technologies underpin countless applications: voice assistants (Siri, Alexa), real-time translation, dictation software, interactive voice response (IVR) systems, and accessibility tools for those with visual or motor impairments. Speech-to-text can dramatically improve productivity, while text-to-speech enables the visually impaired to access written content.
口语与听力技术支撑着无数应用：语音助手（Siri、Alexa）、实时翻译、听写软件、交互式语音应答（IVR）系统，以及为视障或行动障碍者提供的辅助工具。语音转文字可大幅提升生产力，而文字转语音使视障人士能访问书面内容。
However, these advances raise ethical concerns. Voice deepfakes can impersonate individuals, leading to fraud or misinformation. Always-on microphones in smart devices create privacy risks. CAIE expects you to discuss both benefits and drawbacks, backing your arguments with examples. Consider data storage: voice recordings sent to the cloud are vulnerable to breaches.
然而，这些进步也引发了伦理问题。语音深度伪造可以冒充他人，导致欺诈或错误信息。智能设备中始终开启的麦克风带来了隐私风险。CAIE 期望你讨论益处和弊端，并用例子支持论点。考虑数据存储：发送到云的语音录音易受泄露攻击。
You might be asked to evaluate a scenario where a company uses voice data to train AI – discuss informed consent, anonymisation and data minimisation.
你可能遇到这样的评估场景：某公司使用语音数据训练 AI——需讨论知情同意、匿名化和数据最小化。
9. Common Calculation Questions and Traps | 常见计算题与陷阱
Calculation-based items are almost guaranteed to appear. A typical question: 'Calculate the file size in kilobytes for a 30-second mono recording at 44.1 kHz with 16-bit samples.' Always check if the recording is stereo; if so, multiply by 2. Next, pay close attention to units – the formula gives bits, but the answer may be required in bytes, kB or MB.
计算类题目几乎必考。典型问题：'计算 44.1 kHz、16 位采样、30 秒单声道录音的文件大小（千字节）。' 务必检查是否为立体声，若是则乘以 2。接着，密切注意单位——公式得出位，但答案可能要求字节、KB 或 MB。
00 × 16 × 30 = 21,168,000 bits → 21,168,000 / 8 = 2,646,000 bytes → ≈ 2584 kB (or 2.52 MB) | 例：44100 × 16 × 30 = 21,168,000 位 → /8 = 2,646,000 字节 → ≈ 2584 kB (或 2.52 MB)
sufficient (≥ 2 × 20 kHz). Some students incorrectly apply the Nyquist theorem in reverse, thinking the highest reproducible frequency equals the sampling rate – it is actually half the sampling rate.
当心那些同时给出采样频率和最高频率的题目。如果信号包含最高达 20 kHz 的频率，32 kHz 的采样率已足够（≥ 2 × 20 kHz）。有些学生错误地反向应用奈奎斯特定理，以为可重现的最高频率等于采样率——实际上它是采样率的一半。
Also, do not confuse ‘sample rate’ with ‘ monly used in streaming. A question may ask for the bit rate required to stream CD-quality stereo audio in real time.
此外，不要混淆'采样率'和'比特率'。比特率（bps）= 采样率 × 位深度 × 声道数。这在流媒体中常用。题目可能要求计算实时流传输 CD 品质立体声所需的比特率。
10. Linking to the Wider Syllabus | 链接到更广泛的考纲
This topic feeds directly into several syllabus areas: 1.1 Data Representation (sound), 1.3 Hardware (input/output devices, sensors), 2.2 Communication (buffering, streaming protocols), and even 8.1 Security, Privacy and Data Integrity when discussing voice data. A holistic understanding will serve you well in papers that blend concepts.
本主题直接涉及考纲多个领域：1.1 数据表示（声音）、1.3 硬件（输入/输出设备、传感器）、2.2 通信（缓冲、流媒体协议），甚至在讨论语音数据时涉及 8.1 安全、隐私与数据完整性。全面的理解将使你在融合概念的试卷中受益。
For instance, when describing VoIP, you need to combine sound sampling, compression, packet switching, and the need for buffering to mask jitter. Such synoptic questions test your ability to join dots across modules.
例如，描述 VoIP 时，你需要结合声音采样、压缩、分组交换以及需要缓冲来掩盖抖动。这种综合性题目检验你跨模块融会贯通的能力。
Similarly, the use of speech recognition in autonomous systems ties in with artificial intelligence and sensors. Preparing topic links explicitly will elevate your answer quality.
同样地，语音识别在自主系统中的使用与人工智能和传感器联系起来。明确地准备主题关联将提升你的答题质量。
11. Rapid-Fire Revision Checklist | 快速复习清单
- Microphone → ADC → digital samples → CPU / memory (English) | 麦克风 → ADC → 数字样本 → CPU / 内存
- DAC → analogue signal → low-pass filter → speaker (English) | DAC → 模拟信号 → 低通滤波器 → 扬声器
- Nyquist: sampling rate ≥ 2 × fmax (English) | 奈奎斯特：采样率 ≥ 2 × fmax
- File size = f × n × t × channels (bits) (English) | 文件大小 = f × n × t × 声道数（位）
- Speech recognition: acoustic model + language model → text (English) | 语音识别：声学模型 + 语言模型 → 文字
- Text-to-speech: text analysis → prosody prediction → waveform generation (English) | 文语转换：文本分析 → 韵律预测 → 波形生成
- Lossy compression uses psychoacoustic masking (English) | 有损压缩利用心理声学掩蔽
- MIDI stores performance data, not audio waveforms (English) | MIDI 存储演奏数据，而非音频波形
- Privacy concerns with always-on microphones (English) | 始终开启式麦克风的隐私隐忧
12. Exam Technique for Audio Topics | 音频专题的考试技巧
When tackling a 6-8 mark descriptive question, structure your answer logically. Start with the input, move through conversion and storage, and finish with output or application. Use technical vocabulary: transducer, ADC, PCM, quantisation, sample rate, buffer. Explicitly state sequences and show calculations clearly, even if the question only asks for explanation.
应对 6-8 分的描述性题目时，按逻辑组织答案。从输入入手，经过转换和存储，最后以输出或应用结尾。使用技术词汇：换能器、ADC、PCM、量化、采样率、缓冲。明确陈述顺序，即使问题只要求解释，也要清晰展示计算过程。
For comparison questions, create a quick mental table: format, compression type, typical file size, quality, use case. This ensures you cover all assessment objectives.
对于比较类题目，在脑中快速列一个表格：格式、压缩类型、典型文件大小、质量、使用场景。这确保覆盖所有评估目标。
Finally, always link back to the scenario given. If the question states 'a voice-controlled home assistant for elderly users', mention reliability in noisy environments, simple wake words, and privacy safeguards. Contextualising your answer demonstrates application skills and can lift you to the top band.
最后，务必回扣所给场景。如果问题提到'面向老年用户的声控家庭助手'，则要提及嘈杂环境下的可靠性、简单的唤醒词和隐私保护。将答案情境化可展示应用能力，助你拿下高分。
Published by TutorHao | Computer Science Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图