📚 Voice & Audio Processing Exam Prep for CIE A-Level CS | 口语/听力备考专项
This revision guide is not a language speaking or listening test. It focuses on how computers capture, digitise, transmit, synthesise and recognise voice and sound – a key topic area in the CIE A-Level Computer Science (9618) syllabus. You will explore the principles of audio digitisation, compression, speech technologies and real‑time communication that frequently appear in exam questions.
本备考指南不同于语言类口语听力考试。它聚焦计算机如何采集、数字化、传输、合成和识别语音与声音——这是 CIE A-Level 计算机科学（9618）大纲的重要考点。你将系统学习音频数字化原理、压缩技术、语音处理以及实时通信，这些内容在试题中频繁出现。
1. What is Voice & Audio Processing? | 什么是语音与音频处理？
Voice and audio processing encompasses all techniques used to convert, store, manipulate and transmit sound signals in a computing environment. From a simple microphone recording to sophisticated speech recognition, these technologies rely on digitisation, encoding and decoding.
语音与音频处理涵盖了在计算环境中转换、存储、操作和传输声音信号的所有技术。从简单的麦克风录音到复杂的语音识别，这些技术都依赖于数字化、编码与解码。
In the 9618 syllabus, this topic appears under data representation, multimedia, communication and AI. Understanding how sound is sampled and reconstructed is essential for answering exam questions on file size, bit rate and real‑time transport.
在 9618 大纲中，该主题出现在数据表示、多媒体、通信和人工智能等模块。理解声音如何被采样并重建，对于解答文件大小、比特率以及实时传输的考题至关重要。
2. Sound as an Analogue Wave | 作为模拟波的声波
Sound is an analogue phenomenon: continuous variations in air pressure. To be processed by a digital system, these continuous waves must be converted into discrete numerical values through a process called digitisation.
声音是一种模拟现象：气压的连续变化。为了能被数字系统处理，这些连续波必须通过一种称为数字化的过程转换为离散的数值。
Two main characteristics describe a sound wave: amplitude (loudness) and frequency (pitch). The human ear typically perceives frequencies between 20 Hz and 20 kHz. A high‑quality digital representation must capture these parameters accurately.
描述声波的两大特征是振幅（响度）和频率（音高）。人耳通常能感知 20 Hz 到 20 kHz 的频率。高质量的数字表示必须准确捕获这些参数。
3. Sampling and the Nyquist Theorem | 采样与奈奎斯特定理
Sampling measures the amplitude of an analogue signal at discrete time intervals. The number of samples taken per second is the sampling rate, measured in hertz (Hz). To faithfully reconstruct the original signal, the sampling rate must be at least twice the highest frequency present – this is the Nyquist theorem.
采样是在离散时间间隔上测量模拟信号的振幅。每秒采集的样本数称为采样率，单位为赫兹（Hz）。为了忠实地重建原始信号，采样率必须至少是信号中最高频率的两倍——这就是奈奎斯特定理。
Nyquist rate = 2 × fmax
奈奎斯特速率 = 2 × fmax
For example, CD‑quality audio uses a sampling rate of 44.1 kHz because the upper limit of human hearing is about 20 kHz. Undersampling causes aliasing, where high‑frequency components appear as lower, false frequencies in the output.
例如，CD 品质音频使用 44.1 kHz 的采样率，因为人耳听感上限约为 20 kHz。采样不足会导致混叠，高频成分在输出中表现为更低的虚假频率。
4. Quantisation and Bit Depth | 量化与位深度
Quantisation converts each sampled amplitude into a binary number of fixed length. The number of bits used per sample is the bit depth. A higher bit depth allows more precise amplitude representation, reducing quantisation error (noise).
量化把每个采样幅值转换为固定长度的二进制数。每个样本使用的比特数就是位深度。更高的位深度能更精确地表示振幅，减小量化误差（噪声）。
Common bit depths are 8‑bit (256 levels), 16‑bit (65,536 levels) and 24‑bit (over 16 million levels). The CIE specification expects you to calculate the number of quantisation levels using 2n, where n is the bit depth.
常见的位深度有 8 位（256 个等级）、16 位（65,536 个等级）和 24 位（超过一千六百万个等级）。CIE 大纲要求你利用 2n 计算量化等级数，其中 n 是位深度。
5. Calculating Audio File Size | 计算音频文件大小
One of the most common exam tasks is calculating the size of an uncompressed audio file. The formula combines sampling rate, bit depth, duration and number of channels.
最常见的考试任务之一是计算未压缩音频文件的大小。该公式结合了采样率、位深度、时长和声道数。
File size (bits) = Sampling rate (Hz) × Bit depth × Duration (s) × Number of channels
文件大小（比特）= 采样率（Hz）× 位深度 × 时长（秒）× 声道数
To express the result in bytes, divide by 8. For stereo, the number of channels is 2. Always show your working and pay attention to unit conversions (e.g., kilobytes vs kibibytes, though CIE typically uses powers of 10 for simplicity).
要得到以字节为单位的结果，除以 8。对于立体声，声道数为 2。务必展示计算过程，并注意单位换算（如千字节与千比字节，不过 CIE 为简便常用 10 的幂）。
6. Audio Compression Techniques | 音频压缩技术
Compression reduces file size and transmission bandwidth. The two broad categories are lossless and lossy compression. CIE questions often ask you to compare them in the context of audio streaming.
压缩能减小文件大小和传输带宽。两大类是无损压缩和有损压缩。CIE 试题常要求你在音频流媒体背景下比较二者。
Lossless compression (e.g., FLAC) preserves every bit of the original data. It exploits statistical redundancy but offers only modest size reduction. Lossy compression (e.g., MP3, AAC) removes perceptually irrelevant sounds based on psychoacoustic models, achieving much smaller files.
无损压缩（例如 FLAC）保留原始数据的每一位。它利用统计冗余，但仅能适度减小文件。有损压缩（例如 MP3、AAC）根据心理声学模型去除感知上不相关的声音，能极大地缩小文件。
| Feature / 特点 | Lossless / 无损 | Lossy / 有损 |
|---|---|---|
| Data integrity / 数据完整性 | Perfect reconstruction / 完美重建 | Some data discarded / 部分数据丢弃 |
| File size reduction / 文件体积缩减 | Moderate (~30‑50%) / 中等（约30‑50%） | High (up to 90%) / 高（可达90%） |
| Typical use / 典型应用 | Archiving, studio editing / 存档、录音室编辑 | Streaming, portable players / 流媒体、便携播放器 |
7. Speech Synthesis (Text-to-Speech) | 语音合成（文本转语音）
Text‑to‑speech (TTS) systems convert written text into spoken words. They are used in assistive technologies, virtual assistants and navigation systems. A TTS engine typically performs text analysis, prosody prediction and waveform generation.
文本转语音（TTS）系统将书面文本转换为口头语言。它们用于辅助技术、虚拟助手和导航系统。TTS 引擎通常执行文本分析、韵律预测和波形生成。
Modern concatenative synthesis joins pre‑recorded speech fragments, while parametric synthesis generates waveforms from acoustic models. CIE may ask you to explain the limitations of early synthesised voices (robotic, unnatural intonation) and how they have improved.
现代拼接合成拼接预录的语音片段，而参数合成则从声学模型生成波形。CIE 可能会要求你解释早期合成语音的局限性（机械、不自然的语调）以及改进之处。
8. Speech Recognition (Automatic Speech Recognition) | 语音识别（自动语音识别）
Speech recognition maps an acoustic signal to a sequence of words. The process involves feature extraction (e.g., Mel‑frequency cepstral coefficients), acoustic modelling and language modelling. Hidden Markov Models and deep neural networks are commonly used.
语音识别将声学信号映射到单词序列。该过程涉及特征提取（如梅尔频率倒谱系数）、声学建模和语言建模。常用隐马尔可夫模型和深度神经网络。
Challenges include background noise, accents and homophones. CIE expects you to describe how training data and context improve accuracy, and to discuss real‑world applications like voice‑controlled interfaces and dictation software.
挑战包括背景噪声、口音和同音词。CIE 期望你描述训练数据和上下文如何提高准确率，并讨论声控界面和听写软件等实际应用。
9. Voice over IP (VoIP) | IP语音传输
VoIP transmits voice conversations over packet‑switched networks, such as the internet. Audio is digitised, compressed, packetised and delivered using protocols like RTP (Real‑time Transport Protocol) and SIP (Session Initiation Protocol).
VoIP 通过互联网等分组交换网络传输语音对话。音频经数字化、压缩、打包后，使用 RTP（实时传输协议）和 SIP（会话发起协议）等协议进行传送。
Common exam questions focus on the need for low latency, jitter buffering and packet loss concealment. You should also know how VoIP differs from traditional circuit‑switched telephony.
常见的考题关注低延迟需求、抖动缓冲和丢包隐藏。你还需要了解 VoIP 与传统电路交换电话的不同之处。
10. Error Detection and Concealment in Audio Streams | 音频流中的错误检测与隐藏
Real‑time audio streams are vulnerable to packet loss and corruption. Simple error detection techniques like parity bits or checksums flag corrupt packets, but retransmission is often too slow for live conversations.
实时音频流容易发生丢包和数据损坏。简单的错误检测技术如奇偶位或校验和可以标记损坏的封包，但对于实时对话来说，重传通常太慢。
Instead, receivers use concealment strategies: inserting silence, replaying the last valid frame, or interpolating missing samples. You should be able to evaluate these methods in terms of perceived quality.
因此，接收端使用隐藏策略：插入静音、回放上一个有效帧，或插值缺失的采样。你应该能从感知质量的角度评价这些方法。
11. Key Concepts and Exam Tips | 核心概念与应试技巧
Make sure you can define and differentiate: sampling rate, bit depth, bit rate, sampling resolution and quantisation level. Practice file‑size calculations with different durations and stereo/mono settings.
确保你能定义并区分：采样率、位深度、比特率、采样分辨率和量化等级。练习不同时长、立体声/单声道设置下的文件大小计算。
When comparing VoIP with PSTN, mention bandwidth efficiency, cost and the challenges of latency and jitter. Use diagrams to illustrate the ADC (analogue‑to‑digital conversion) process. Always link compression methods to their impact on quality and file size.
在比较 VoIP 与 PSTN 时，提及带宽效率、成本以及延迟和抖动的挑战。使用图表说明模数转换（ADC）过程。始终将压缩方法与其对质量和文件大小的影响联系起来。
12. Conclusion | 结语
Voice and audio processing spans multiple areas of the CIE A‑Level Computer Science syllabus, from fundamental data representation to advanced communication and AI. Mastering these concepts not only prepares you for the exam but also gives you insight into technologies that shape everyday life, from smart assistants to online conferencing.
语音与音频处理横跨 CIE A-Level 计算机科学大纲的多个领域，从基础数据表示到高级通信和人工智能。掌握这些概念不仅为考试做好准备，还能让你深入理解从智能助手到在线会议等塑造日常生活的技术。
Regularly review calculations, terminology and real‑world scenarios. With a strong grasp of digitisation, compression and real‑time protocols, you will be able to tackle any voice or audio question with confidence.
定期复习计算、术语和实际场景。扎实掌握数字化、压缩和实时协议后，你就能自信地应对任何语音或音频相关的问题。
Published by TutorHao | Computer Science Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
屏轩国际教育cambridge primary/secondary checkpoint, cat4, ukiset,ukcat,igcse,alevel,PAT,STEP,MAT, ibdp,ap,ssat,sat,sat2课程辅导，国外大学本科硕士研究生博士课程论文辅导