📚 Year 10 Eduqas Computer Science: Speech & Audio Technology Revision | 口语/听力备考专项
This revision guide unpacks the computer science behind speaking and listening. From capturing analogue sound waves to synthesising human-like speech, we cover the Eduqas GCSE topics of sound representation, sampling, speech recognition, and audio processing. Whether you are preparing for an exam or deepening your understanding of voice technology, these explanations and calculations will build your confidence.
本复习指南剖析计算机“听说”背后隐藏的学问。从采集模拟声波到合成逼真语音，我们系统梳理Eduqas GCSE关于声音表示、采样、语音识别与音频处理的核心考点。不论你正在备考冲刺，还是想彻底弄懂语音技术，这里的概念解析与计算演练都能帮你牢固掌握。
1. Sound as an Analogue Signal | 作为模拟信号的声音
Sound is created by vibrations that travel through the air as continuous pressure waves. A microphone converts these pressure variations into a continuously changing electrical voltage, which is an analogue signal. Unlike digital data, an analogue signal has an infinite number of possible values at any moment in time.
声音由振动产生，以连续压力波的形式在空气中传播。麦克风将这些压力变化转换为持续变化的电压，形成模拟信号。与数字数据不同，模拟信号在任何瞬间都有无限多种可能的取值。
Computers cannot store or process infinite precision directly. Therefore, an analogue-to-digital conversion (ADC) process must sample the signal at discrete intervals and map each sample to a fixed digital value. This process lies at the heart of all digital audio systems, from voice memos to music streaming.
计算机无法直接处理无限精度的信号。因此，必须依靠模数转换(ADC)过程，以离散的时间间隔对信号采样，并将每个样本映射为固定的数字值。从语音备忘录到流媒体音乐，这一过程是所有数字音频系统的基石。
2. Sampling and Sample Rate | 采样与采样率
Sampling is the process of measuring the amplitude of an analogue signal at regular time intervals. The sample rate, measured in hertz (Hz), tells us how many samples are taken per second. A higher sample rate captures more detail and can reproduce higher frequencies more accurately.
采样是指以固定时间间隔测量模拟信号幅值的过程。采样率以赫兹(Hz)为单位，表示每秒采集的样本数量。更高的采样率能捕获更多细节，更精确地重现高频声音。
Common sample rates include 8,000 Hz for telephone voice, 44,100 Hz for CD-quality audio, and 48,000 Hz or 96,000 Hz for professional recording. When the sample rate is too low, high-frequency components are lost, and the reconstructed sound may sound muffled or distorted.
常见的采样率包括电话语音的8000 Hz、CD音质的44100 Hz以及专业录音的48000 Hz或96000 Hz。当采样率过低时，高频信息丢失，还原的声音听起来会发闷或失真。
The choice of sample rate involves a trade-off between audio fidelity and file size. For speech-only applications like voice assistants, lower sample rates save storage and bandwidth without significantly harming intelligibility.
采样率的选择要在音频保真度和文件大小之间权衡。对于语音助手这类纯语音应用，较低的采样率可节省存储和带宽，对可懂度影响不大。
3. Bit Depth and Quantisation | 比特深度与量化
nes how many discrete amplitude levels are available. For example, 8-bit audio provides 2⁸ = 256 levels, while 16-bit audio offers 2¹⁶ = 65,536 levels.
采集样本后，其幅值必须被归约到有限集合中最接近的层级，这一过程称为量化。比特深度决定了可用的离散幅值层级数量。例如，8位音频提供256个层级，16位音频则提供65536个层级。
Higher bit depth reduces quantisation error, also known as quantisation noise, giving a cleaner, more dynamic recording. CD audio uses 16 bits per sample; professional studio work often uses 24 bits. Telephone systems may use only 8 bits, balancing quality with data rate.
更高的比特深度能减小量化误差（也称量化噪声），呈现出更纯净、动态范围更广的录音。CD音频使用16位每样本；专业录音室常用24位。电话系统可能只使用8位，在质量与数据速率之间取得平衡。
Bit depth directly impacts the signal-to-noise ratio. A rough rule of thumb is that each extra bit improves the signal-to-noise ratio by about 6 dB. Together, sample rate and bit depth define the resolution and fidelity of digital audio.
比特深度直接影响信噪比。一个粗略的经验法则是，每增加1位，信噪比约提升6 dB。采样率与比特深度共同决定了数字音频的分辨率与保真度。
4. File Size Calculation for Audio | 音频文件大小计算
Uncompressed audio file size can be calculated using the fundamental formula:
File size (bits) = sample rate (Hz) × bit depth × duration (seconds) × number of channels
无压缩音频文件大小可用基本公式计算：
文件大小(bits) = 采样率(Hz) × 比特深度 × 时长(秒) × 声道数
bytes by dividing by 8 to get bytes, then by 1,024 twice. This yields approximately 10.1 MB.
例如，一段CD音质立体声录音（44100 Hz，16位，双声道）1分钟：44100 × 16 × 60 × 2 = 84672000 bits。转换为兆字节时，先除以8得到字节，再除以1024两次，结果约为10.1 MB。
In exams, you may be asked to calculate file sizes and compare data requirements across different sampling settings. Always show your working step by step, clearly stating the units at each stage.
考试中常会要求计算文件大小，并比较不同采样设置下的数据需求。务必逐步展示计算过程，并在每一步明确注明单位。
5. The Nyquist Theorem | 奈奎斯特定理
The Nyquist theorem states that to faithfully reconstruct a signal, the sample rate must be at least twice the highest frequency present in the signal. In symbols, fs ≥ 2 × fmax.
奈奎斯特定理指出，要无失真地重建信号，采样率必须至少是信号中最高频率的两倍。记作 fs ≥ 2 × fmax。
Human hearing typically ranges up to about 20,000 Hz. Therefore, the CD sample rate of 44,100 Hz satisfies the Nyquist requirement, allowing frequencies up to 22,050 Hz to be captured. Telephone speech is limited to roughly 3,400 Hz, so an 8,000 Hz sample rate is sufficient.
人耳听觉范围通常可达约20000 Hz。因此，CD的44100 Hz采样率满足奈奎斯特要求，可捕获的最高频率达22050 Hz。电话语音的上限约3400 Hz，所以8000 Hz的采样率已足够。
If the sample rate is too low, aliasing occurs — high-frequency components are misrepresented as lower frequencies, creating unpleasant artefacts. An anti-aliasing filter is applied before sampling to remove frequencies above half the sample rate.
若采样率过低，则发生混叠——高频分量被错误地表现为低频，产生刺耳失真。采样前会使用抗混叠滤波器，滤除高于采样率一半的频率。
6. Analogue-to-Digital Conversion (ADC) | 模数转换器
An ADC integrates sampling and quantisation to turn continuous analogue voltages into discrete binary numbers. The process typically follows three stages: anti-aliasing filtering, sample-and-hold, and quantisation. Each sample is held steady while the converter assigns a binary code.
模数转换器(ADC)整合采样与量化，将连续变化的模拟电压转变为离散的二进制数。过程通常分为三个阶段：抗混叠滤波、采样保持和量化。每个样本被保持稳定，同时转换器为其分配二进制编码。
The number of bits output by the ADC equals its bit depth. For an 8-bit ADC, each sample is represented by an 8-bit binary number. The speed of the ADC determines the maximum possible sample rate, and its accuracy is measured by parameters such as integral linearity.
ADC输出的位数即为其比特深度。对于8位ADC，每个样本由一个8位二进制数表示。ADC的速度决定了可达到的最高采样率，其精度由积分线性度等参数衡量。
Modern sound cards and smartphone components contain high-quality ADCs. Understanding ADC principles helps you appreciate why hardware choices affect recording quality and why compressed formats are so prevalent.
现代声卡和智能手机组件都集成了高品质ADC。理解ADC的工作原理，有助于你明白为何硬件选择会影响录音质量，以及压缩格式为何如此普及。
7. Introduction to Speech Recognition | 语音识别简介
Speech recognition, often called automatic speech recognition (ASR), enables a computer to convert spoken words into text. The process involves capturing audio, extracting acoustic features, and applying statistical or neural models to decode the most likely word sequence.
语音识别，常称为自动语音识别(ASR)，让计算机能将口述词语转换为文字。过程包括采集音频、提取声学特征，并应用统计或神经模型解码出最可能的词序列。
Feature extraction usually involves computing Mel-Frequency Cepstral Coefficients (MFCCs), which represent short-term power spectrum shapes modelled on human ear perception. These feature vectors are then fed into an acoustic model, a pronunciation dictionary, and a language model.
特征提取通常包括计算梅尔频率倒谱系数(MFCC)，这些系数模仿人耳感知特性来表示短时功率谱形状。随后，特征向量被送入声学模型、发音词典和语言模型。
Modern ASR systems, such as those in smartphones and smart speakers, use deep neural networks trained on vast speech corpora. Despite impressive accuracy, challenges remain with accents, background noise, and homophones.
智能手机和智能音箱中的现代ASR系统采用在大型语音语料库上训练的深度神经网络。尽管准确率惊人，但在口音、背景噪音和同音词方面仍面临挑战。
8. Text-to-Speech Synthesis | 文本到语音合成
Text-to-speech (TTS) does the reverse of speech recognition: it converts written text into spoken audio. Early TTS systems used concatenative synthesis, stitching together pre-recorded speech fragments. Modern approaches employ parametric and neural models to generate more natural prosody.
文本到语音(TTS)完成与语音识别相反的任务：将书面文字转化为口语。早期TTS使用拼接合成，把预录语音片段拼接起来。现代方法则使用参数化模型和神经模型，生成更自然的韵律。
A TTS pipeline typically includes text normalisation, linguistic feature extraction, and waveform generation. Neural TTS models like Tacotron and WaveNet directly predict waveforms from text, producing remarkably human-like speech with appropriate intonation and pacing.
TTS流程通常包括文本规范化、语言特征提取和波形生成。像Tacotron和WaveNet这类神经TTS模型直接从文字预测波形，能够生成音调和节奏适宜、极其接近人声的语音。
Applications range from screen readers for the visually impaired to virtual assistants and navigation systems. The balance between naturalness and computational efficiency remains an active area of research.
应用场景涵盖视障人士的屏幕朗读器、虚拟助手和导航系统。自然度与计算效率之间的平衡仍是活跃的研究领域。
9. Audio Compression Techniques | 音频压缩技术
Raw audio files are large, so compression is essential. Compression can be lossless or lossy. Lossless compression (e.g., FLAC, ALAC) reduces file size without discarding any audio information, exploiting statistical redundancy. Lossy compression (e.g., MP3, AAC) removes sounds less perceptible to the human ear to achieve much smaller sizes.
原始音频文件巨大，因此压缩必不可少。压缩分为无损和有损两类。无损压缩（如FLAC、ALAC）在不去弃任何音频信息的前提下，利用统计冗余缩减文件体积。有损压缩（如MP3、AAC）则移除人耳较不易感知的声音，从而大幅缩小文件。
MP3 encoding uses psychoacoustic models to determine which frequency components can be masked or discarded. Bit rate, measured in kilobits per second (kbps), controls the amount of compression — 128 kbps is a common trade-off, while 320 kbps preserves more detail.
MP3编码借助心理声学模型，判定哪些频率分量可以被掩蔽或丢弃。比特率以千比特每秒(kbps)为单位，控制压缩程度——128 kbps是常见的折衷选择，320 kbps则保留更多细节。
| Format | Type | Typical Use |
|---|---|---|
| WAV | Uncompressed | Master recording |
| FLAC | Lossless | Archiving, hi-fi audio |
| MP3 | Lossy | Streaming, portable players |
| AAC | Lossy | YouTube, Apple Music |
When comparing compression methods, consider both file size reduction and perceived quality. Lossy formats are suitable for casual listening and streaming, while lossless formats are preferred for editing and archival.
比较压缩方法时，需兼顾文件体积缩减与主观听觉质量。有损格式适合日常聆听和流媒体，无损格式则更适合编辑和存档。
10. Ethics and Applications of Voice Technology | 语音技术的伦理与应用
Voice technology powers virtual assistants, hands-free controls, accessibility tools, and real-time translation. However, it also raises ethical concerns around privacy, data security, and consent. Smart speakers continuously listen for wake words, and voice data is often uploaded to cloud servers for processing.
语音技术驱动着虚拟助手、免提控制、无障碍工具和实时翻译，但同时也引发了关于隐私、数据安全和用户同意的伦理问题。智能音箱持续监听唤醒词，语音数据通常上传至云端处理。
Storing and analysing voice samples can reveal sensitive information such as identity, emotional state, and health conditions. Legislation like GDPR requires companies to be transparent about data usage. As computer scientists, we must design systems that respect user rights and minimise unnecessary data collection.
存储和分析语音样本可能暴露身份、情绪状态和健康状况等敏感信息。GDPR等法规要求企业对数据用途保持透明。作为计算机科学的学习者和从业者，我们必须设计尊重用户权利、减少不必要数据收集的系统。
Potential biases in speech recognition models can lead to poorer accuracy for certain accents, dialects, or background noises. Ethical AI development calls for diverse training data and continuous fairness testing.
语音识别模型中的潜在偏见可能导致对某些口音、方言或背景噪音的准确率偏低。负责任的AI开发需要多样化的训练数据，以及持续的公平性测试。
11. Key Revision Points Summary | 复习要点速览
To wrap up, make sure you can define sampling, sample rate, bit depth, and quantisation. Practise calculating uncompressed audio file sizes and explain why the Nyquist theorem matters. Understand the basic pipeline of speech recognition and text-to-speech, and list advantages and drawbacks of lossy vs lossless compression.
总结一下，务必能定义采样、采样率、比特深度和量化；熟练计算无压缩音频文件大小，并解释奈奎斯特定理的意义；理解语音识别和文本到语音的基本流程，列出有损与无损压缩的优缺点。
Review the ethical dimensions, especially around data privacy and algorithmic fairness. Revisit real-world applications to help you link theory to practical examples. Test yourself by setting up short comparison tables and working through past paper questions on sound representation.
复习伦理维度，特别是数据隐私和算法公平性。回顾现实应用，帮助你把理论与实际案例联系起来。可通过制作简要对比表和练习历年真题中的声音表示题目来自我检测。
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