审阅范围 · Review scope
公开来源文字与原创综合已审阅。请保留正文中的出处与限制；不代表全库最终验收、实际演奏或学习效果验证。
Public-source text and original synthesis reviewed. Keep the evidence and limitations stated in the text in view. This is not final corpus acceptance or validation of performance or learning outcomes.
第十七批：音訊製作與遠端學習 / Batch 17: Audio production and remote learning
六篇公開正文：效果器、修復、編曲與遠端教學；以原聲鋼琴與紙筆為預設。
Six public article bodies on effects, restoration, arranging, and remote teaching; acoustic piano and paper are the default.
範圍與限制 / Scope and limits
來源 ID、指派網址與中文原題完整保留；英文題名為編輯翻譯。Markdown 與 JSON 是同一成果，不可重複計算。 Source IDs, assigned URLs, and original Chinese titles are retained exactly; English titles are editorial translations. Markdown and JSON represent one deliverable, not two reviews.
讀到頁面文字不等於聽過示範、看過圖像或測過軟體。主要文章媒體、連結課程與會員內容均未檢驗，亦不補寫其內容。 Reading page text is not hearing demonstrations, inspecting images, or testing software. Primary-article media, linked courses, and member content remain unreviewed and are not reconstructed.
文章中的版本、平台、價格、介面、網路速度、安全性與主觀品質比較只屬發表當時敘述；不當作目前購買、安裝或部署建議。補充文件也只驗證其文字，未驗證實際產品行為。 Article versions, platforms, prices, interfaces, network speed, security, and subjective comparisons are historical accounts, not current purchase, installation, or deployment advice. Supplementary documents verify text only, not actual product behavior.
每個來源的雙語來源衍生文字均計入題名、摘要、用途、證據、涵蓋、限制及任何跨段重述；以中文字元加英數詞元保守計算，合計最多 200。原創數值題不冒充作者教材。 Each source-derived budget includes both-language titles, summaries, roles, evidence, coverage, limits, and cross-section restatements. Conservative CJK-character plus alphanumeric-token counting is capped at 200. Original numerical exercises are not presented as author teaching materials.
所有練習都有完整紙上材料、步驟、參考答案及停止條件；先遮住答案，寫下預測再核對。已看過答案就標記已見，不能聲稱是盲測或學習增益。 Every exercise supplies paper materials, steps, answer guidance, and a stopping condition. Cover answers and commit predictions first. Mark previously seen answers as seen; do not claim blinded testing or learning gains.
不需唱歌、錄音、電子鋼琴、外接設備、帳號或軟體。可選試彈時用現有原聲鋼琴，自選舒適拍速；不舒服或不熟鍵位就只做紙上題。 No singing, recording, digital piano, external equipment, account, or software is needed. Optional playing uses the existing acoustic piano at a comfortable pace; use paper only if playing is uncomfortable or key locations are unfamiliar.
可選試彈統一約定：右手一次一音，左手休息，不踩踏板；C4 是中央 C，往右相鄰白鍵為 D4、E4、F4、G4。四音例均用 4/4 一小節、每音一拍；第四卡保留兩小節。自選舒適速度，不必唱或錄音。 Shared optional-playing convention: right hand, one note at a time; left hand rests, with no pedal. C4 is middle C; successive white keys right are D4, E4, F4, G4. Four-note examples occupy one 4/4 bar, one beat per note; card four retains two bars. Choose a comfortable pace, with no singing or recording.
本批未安裝、購買、開帳號、修改網路、授予麥克風權限、啟動音訊應用、上傳來源或推送；未啟動本機模型。未觀察到存取拒絕或用量耗盡。 This batch performed no installation, purchase, account creation, network change, microphone permission, audio-app launch, source upload, or push, and started no local model. No access denial or usage exhaustion was observed.
藝術家名稱只定位歷史案例；不複製旋律、歌詞、樂譜或編曲分軌，不把單一配方說成藝術家的全貌。 Artist names contextualize a historical case only. No melody, lyrics, score, or arrangement stems are copied, and no single recipe is treated as defining an artist.
原創整合判斷 / Original synthesis
把資料分成四層：演奏事件、房間內聲音、記錄或傳輸結果、後製版本。回饋先註明證據在哪一層，再問是否足以談音符、拍點、觸鍵或踏板。 Separate four layers: performance events, room sound, recorded/transmitted output, and edited versions. Feedback first identifies the evidence layer, then asks whether it supports claims about notes, timing, touch, or pedal.
未加工檔也不是無誤的現場真相；它仍經過收音與編碼。保留它的目的，是保留比較基準與修改履歷，不是替它背書。 An unedited file is not infallible live truth; capture and encoding still intervene. Retaining it preserves a comparison baseline and edit history, not a guarantee of accuracy.
看到怪聲先列競爭解釋，不先指責學生。可以記為未確定；不能由單一波形或修復版本推論手指姿勢、用力過大、踏板錯誤或身體風險。 List competing explanations for odd sound before blaming the learner. Unresolved is a valid outcome; one waveform or restored version cannot establish finger posture, excessive force, pedal mistakes, or bodily risk.
本批數值與事件表全為原創假想；只檢查推理與算術。沒有真實聲音、連線或學習者表現可供測量，不能宣稱聽感、準確率或教學效果已成立。 All numerical and event tables are original hypotheticals that test reasoning and arithmetic only. No real sound, connection, or learner performance was measured; audibility, accuracy rates, and teaching efficacy are not established.
1. 好和弦的混音幼幼班 - Chorus/Flanger/Phaser 篇 / Mixing for beginners: Chorus, Flanger, and Phaser
來源 / Source: WEB-58fbe760bc5a · 原文 / Primary article
發表 / Published: 2021-10-15 · 閱讀 / Read: 2026-10-22
涵蓋 / Coverage: 30–91 / 94 lines
完整正文；媒體未審。
Complete body; media unreviewed.
來源摘要 / Source summary
Chorus 疊加微變時間與音高的副本；Flanger 混入變動短延遲；Phaser 混入全通濾波改變相位的支路。
Chorus layers slightly time/pitch-varied copies; flanger mixes a changing short delay; phaser mixes an allpass phase-shifted branch.
內容角色 / Content role
辨別處理機制。
Distinguish processing mechanisms.
證據類型 / Evidence type
作者教學。
Author exposition.
限制 / Limits
聽感比喻不是唯一識別法；未聽示範。
Timbre metaphors are not unique identifiers; demonstrations unheard.
來源衍生雙語預算 / Source-derived bilingual budget: 125/200; 正文逐字引述 / body quotation: 0.
原創應用（非作者教材）/ Original application (not author teaching material)
練習 / Exercise
同一次演奏，三個可能的檔案
One performance, three possible files
目的 / Purpose
練習先找訊號履歷，再決定能否把變化算成演奏錯誤；不訓練效果器聽辨。
Practice checking signal history before treating a change as a playing error; this is not effect ear-training.
材料 / Materials
紙、筆與下列三張虛構紀錄；不需要音檔。數字是任意刻度，不是分貝或鋼琴力度。
Paper, pencil, and the three fictional records below; no audio file is required. Numbers are arbitrary units, not decibels or piano dynamics.
完整資料卡 / Complete data card
名詞 / Terms
本題 x 是輸入數列，z 是處理支路，y 是兩者相加的結果。相位可先想成週期內走到哪個位置；正負數代表方向而非好壞。
Here x is an input sequence, z a processed branch, and y their sum. Think of phase as position within a cycle; positive and negative indicate direction, not quality.
紀錄 / Records
1.
識別 / ID
A
描述 / Description
只有演奏者的紙上記錄：C4、D4、E4、C4，各一拍；沒有聲音。
Only the player’s written log: C4, D4, E4, C4, one beat each; no sound.
2.
識別 / ID
B
描述 / Description
標示由同一次取樣而來的加工檔，沒有可查處理履歷；有人說聽到音高擺動。此說法未經核實。
An edited file labeled as coming from that same take has no inspectable processing history; someone reports pitch movement. The report is unverified.
3.
識別 / ID
C
描述 / Description
純算術玩具：x=[1,0,-1,0]；z₁=[1,0,-1,0]；z₂=[-1,0,1,0]。
Arithmetic toy only: x=[1,0,-1,0]; z₁=[1,0,-1,0]; z₂=[-1,0,1,0].
任務 / Tasks
計算 y₁=x+z₁、y₂=x+z₂；再判斷 B 是否足以證明學生彈錯音。
Calculate y₁=x+z₁ and y₂=x+z₂; then decide whether B proves that the learner played a wrong note.
步驟 / Steps
先寫兩個預測：哪個加總會抵消？B 還缺什麼證據？不能把未核實的描述當成已聽見。 First predict which sum cancels and what evidence B lacks. Do not treat an unverified description as something you heard.
逐欄相加。這只是對齊或反向的數列示例，並非完整的動態效果器，也不是從文章截取的波形。 Add column by column. These aligned/opposed sequences are not complete time-varying effects or waveforms copied from the article.
回到摘要，分別給三個機制標籤；保留「由音檔外觀單獨不能分類」選項，勿由比喻猜演算法。 Return to the summary and label the three mechanisms separately. Keep “not classifiable from file appearance alone” as an option rather than guessing an algorithm from a metaphor.
可選在原聲鋼琴單手依 A 的四音慢彈一次；C4 是中央 C，D4 是其右方相鄰白鍵，E4 再往右一個白鍵。其後只寫自己的觀察，不假裝製造了數位效果。 Optionally play A slowly with one hand on the acoustic piano: C4 is middle C, D4 the next white key right, and E4 the next one. Write your own observation afterward, without pretending to have generated digital effects.
參考答案 / Answer guidance
y₁=[2,0,-2,0]；y₂=[0,0,0,0]。兩個結果都來自相同 x；改變輸出不必代表輸入事件改變。 y₁=[2,0,-2,0]; y₂=[0,0,0,0]. Both use the same x; changed output need not imply changed input events.
B 不足以判定錯音；需要可查履歷與合適的比較證據。A 也只是自述，不能取代實際演奏證據。 B cannot establish a wrong note; inspectable history and suitable comparison evidence are needed. A is also a self-report, not a substitute for performance evidence.
完成條件 / Completion check
兩個向量正確，三個名詞不混為一談，並在證據不足時保留未知，即可停止。
Stop after obtaining both vectors, keeping the three terms distinct, and preserving uncertainty under insufficient evidence.
已揭答案變體 / Answer-disclosed variant
改用 x=[2,-2]、z=[-1,1]，答案是 [1,-1]；這是已揭答案的練習，不是盲測。
Use x=[2,-2] and z=[-1,1]; the result is [1,-1]. This answer-disclosed variant is practice, not a blinded test.
2. 好和弦的混音幼幼班 - 破音篇 / Mixing for beginners: Distortion
來源 / Source: WEB-e89c4759c5a2 · 原文 / Primary article
發表 / Published: 2021-10-24 · 閱讀 / Read: 2026-10-22
涵蓋 / Coverage: 30–143 / 146 lines
完整正文；媒體未審。
Complete body; media unreviewed.
來源摘要 / Source summary
以單頻正弦波觀察 Overdrive、Distortion、Fuzz 的波形與諧波（基頻整數倍）變化，示例從輕微彎曲到強烈變形，並談藝術用途。
Uses a single-frequency sine wave to examine waveform and harmonic changes from Overdrive, Distortion, and Fuzz. Harmonics are integer multiples of the fundamental. Examples range from mild bending to severe reshaping; artistic uses follow.
內容角色 / Content role
分開音色與錯誤。
Separate timbre from error.
證據類型 / Evidence type
作者教學。
Author exposition.
限制 / Limits
泛音分布屬該 Scream 4 示例，非通用分類；不模仿損壞器材。
Harmonic patterns belong to the Scream 4 examples, not a universal taxonomy; do not imitate equipment damage.
來源衍生雙語預算 / Source-derived bilingual budget: 161/200; 正文逐字引述 / body quotation: 0.
原創應用（非作者教材）/ Original application (not author teaching material)
練習 / Exercise
多個輸入，同一輸出
Many inputs, one output
目的 / Purpose
用自己定義的數學轉換理解資訊可失去，避免把破音直接解讀成彈得太用力。
Use explicitly defined mathematical transformations to see how information can be lost, without interpreting distortion directly as excessive playing force.
材料 / Materials
紙與筆；以下函數是本題的定義，不是任何產品模型或設備設定。
Paper and pencil; the functions below are definitions for this exercise, not product models or equipment settings.
完整資料卡 / Complete data card
輸入 x / Input x
-2-0.500.52
定義 / Definitions
A(x)=x/2；B(x)：低於 -1 取 -1，高於 1 取 1，其他保留 x。x 沒有物理單位；A 是比例縮放，B 是本題的截頂規則。
A(x)=x/2. B(x) returns -1 below -1, 1 above 1, and otherwise x. Values have no physical unit; A is proportional scaling, and B is this exercise’s clipping rule.
問題 / Questions
計算兩列；另算 B(2)、B(3)。只知道 B(x)=1 能唯一找回 x 嗎？
Calculate both rows and also B(2), B(3). Does knowing only B(x)=1 uniquely recover x?
步驟 / Steps
先預測每個輸出，再計算；不要把數值 2 解讀成某一琴鍵速度或聲壓。 Predict each output, then calculate; do not interpret 2 as a particular key velocity or sound pressure.
列出至少兩個能得到同一輸出的輸入；這是反推不唯一的證明，不需要聽聲音。 List at least two inputs yielding the same output. This proves non-unique inversion without listening.
在紙上分開兩種目的：創作版本可接受刻意改變；演奏評量要知道改變發生在哪裡。兩者不可共用未註明版本的分數。 Separate two goals on paper: a creative version may deliberately change sound; performance assessment needs the location of changes. They must not share a score with no version label.
對虛構評語「這個檔案破了，所以你必定彈太重」寫出保留證據的改寫。此題不需彈奏。 Rewrite the fictional comment “this file distorts, so you definitely played too hard” to preserve evidential limits. No playing is needed.
參考答案 / Answer guidance
A=[-1,-0.25,0,0.25,1]；B=[-1,-0.5,0,0.5,1]；B(2)=B(3)=1，因此只知輸出 1 無法唯一反推輸入。 A=[-1,-0.25,0,0.25,1]; B=[-1,-0.5,0,0.5,1]; B(2)=B(3)=1, so output 1 alone cannot uniquely determine the input.
可改為「檔案被描述有失真；來源與處理鏈尚未核實，暫不對觸鍵下結論。」不能僅由本題函數聲稱已修復錄音。 One rewrite: “The file is reported to contain distortion; capture and processing are unverified, so no touch judgment is warranted yet.” These functions do not establish that any recording was repaired.
完成條件 / Completion check
算對兩列並找出多對一例子；寫出一條不歸咎身體動作的回饋即完成。
Complete both rows, identify a many-to-one example, and write one feedback statement that does not blame bodily technique.
已揭答案變體 / Answer-disclosed variant
將界線改為 ±0.75；x=1 與 x=2 都輸出 0.75。這仍是玩具規則，不是通用修復算法。
Change the bounds to ±0.75: x=1 and x=2 both output 0.75. This remains a toy rule, not a universal restoration algorithm.
3. 拯救檸檬卷的國中畢業歌！如何修復音質超爛的錄音？ / Rescuing a school graduation song: Repairing a poor recording
來源 / Source: WEB-3b6dc964eea8 · 原文 / Primary article
發表 / Published: 2021-07-26 · 閱讀 / Read: 2026-10-22
涵蓋 / Coverage: 30–123 / 126 lines
完整正文；媒體未審。
Complete body; media unreviewed.
來源摘要 / Source summary
案例把降噪、分離、音高／時間編修、混音與新增聲部串成翻修流程。
The case combines denoising, separation, pitch/timing edits, mixing, and added parts into a renovation workflow.
內容角色 / Content role
追蹤版本與證據。
Track versions and evidence.
證據類型 / Evidence type
作者教學。
Author exposition.
限制 / Limits
文中已承認資訊遺失；軟體描述屬 2021 年，翻修不是原始演奏真相。
The article acknowledges lost information; software descriptions are from 2021, and renovation is not original-performance truth.
來源衍生雙語預算 / Source-derived bilingual budget: 145/200; 正文逐字引述 / body quotation: 0.
原創應用（非作者教材）/ Original application (not author teaching material)
練習 / Exercise
修復帳本，不是時間機器
A restoration ledger, not a time machine
目的 / Purpose
區分保存、估計、改寫與新演奏，使「比較好聽」不被當作「原來就是這樣」。
Separate preservation, estimation, rewriting, and new performance so that “more pleasing” does not become “what originally happened.”
材料 / Materials
紙、筆及完整虛構帳本；沒有提供或需要第三方音訊。
Paper, pencil, and the complete fictional ledger; no third-party audio is supplied or needed.
完整資料卡 / Complete data card
列 / Rows
1.
識別 / ID
R0
描述 / Description
封存一份未再編輯的混合記錄；來源、設備與處理前狀態未知。
Archive an otherwise unedited mixed record; provenance, equipment, and earlier processing are unknown.
2.
識別 / ID
R1
描述 / Description
依估計的背景模板減少部分成分。
Reduce some components using an estimated background template.
3.
識別 / ID
R2
描述 / Description
把混合成分拆成兩個估計支路。
Split mixed components into two estimated branches.
4.
識別 / ID
R3
描述 / Description
把原事件的開始位置移到預定格線。
Move original event starts onto a chosen grid.
5.
識別 / ID
R4
描述 / Description
刪掉一段，另寫四個音取代。
Delete one passage and write four replacement notes.
6.
識別 / ID
R5
描述 / Description
輸出較悅耳的版本；沒有獨立原始參考。
Export a more pleasing version without an independent original reference.
玩具混合 / Toy mixture
只知道 x=a+b=10；候選一是 a=6,b=4，候選二是 a=3,b=7。這是任意數值，並非實際頻譜。
Only x=a+b=10 is known; one candidate is a=6,b=4 and another a=3,b=7. These are arbitrary values, not an actual spectrum.
標籤 / Labels
可選標籤：保存、估計、改寫、審美判斷。允許一列多標籤。
Available labels: preservation, estimation, rewriting, aesthetic judgment. Multiple labels per row are allowed.
步驟 / Steps
先遮住答案，把 R0–R5 分類，寫出哪些版本適合比較編輯前後，哪些不能用來替原演奏打分。 Cover the answers and classify R0–R5. Identify comparison versions and those unsuitable for scoring the original performance.
核對兩組加法。若某方法輸出 a=6，僅憑 x=10 能證明它猜對嗎？寫下還需要哪類獨立證據。 Check both sums. If a method outputs a=6, does x=10 alone prove correctness? State the kind of independent evidence needed.
為 R3、R4 各寫一條履歷：改了什麼、為什麼改、不能由結果主張什麼。不要把喜好改寫標成原作者的錯誤。 Write a log for R3 and R4: what changed, why, and what the result cannot establish. Do not label preference-based rewriting as the original author’s error.
可選另創四音 C4、E4、D4、G4，各一拍；只在自己的原聲鋼琴單手慢彈。這是新的示例，與畢業歌無關；不錄音也不評估來源人物。 Optionally use the new four-note example C4, E4, D4, G4, one beat each, slowly with one hand on your acoustic piano. It is unrelated to the graduation song; do not record or evaluate the source’s performers.
參考答案 / Answer guidance
R0 是保存，但不保證現場真相；R1、R2 是估計；R3、R4 是改寫；R5 是審美判斷。R0 可作版本基準，不能自動作演奏真值。 R0 is preservation, not guaranteed live truth; R1 and R2 are estimation; R3 and R4 are rewriting; R5 is aesthetic judgment. R0 is a version baseline, not automatically performance ground truth.
6+4=3+7=10，故 a 的反推不唯一。獨立、可核實的原始分量才可能檢驗估計；「變好聽」本身不足。 6+4=3+7=10, so a is not uniquely determined. Independently verifiable original components could test the estimate; “sounds better” alone cannot.
R3 可標「為新伴奏對齊時間；不得據此聲稱原演奏已如此穩定」；R4 可標「新增音符；不是找回遺失音符的證據」。 R3 may say “timing aligned for a new accompaniment; not evidence that the original was this steady.” R4 may say “new notes added; not evidence of recovered missing notes.”
完成條件 / Completion check
分類六列、列出不唯一分解、留下兩條修改履歷即完成；不能宣稱已修好任何真實錄音。
Finish after classifying six rows, showing non-unique decomposition, and writing two edit logs; do not claim to have repaired any real recording.
回饋規則 / Feedback rule
評量界面應將「來源演奏」「估計重建」「創作改寫」分欄；經改音或改拍版本不能冒充未編輯練習。
An assessment interface should separate source performance, estimated reconstruction, and creative rewriting; pitch/time-edited versions must not masquerade as unedited practice.
4. 如何做出 YOASOBI 風的電子舞曲？ / Making YOASOBI-style electronic dance music
來源 / Source: WEB-cecbddae1eb1 · 原文 / Primary article
發表 / Published: 2023-09-06 · 閱讀 / Read: 2026-10-22
涵蓋 / Coverage: 30–142; 151 / 154 lines
完整正文；媒體未審。
Complete body; media unreviewed.
來源摘要 / Source summary
由樂器清單拆解一個片段，逐層安排鋼琴、節奏與配器，並調整格線與音色。
Breaks down one passage with an instrument inventory, layering piano, rhythm, and arrangement while adjusting grid alignment and timbre.
內容角色 / Content role
從角色規劃編曲。
Plan arrangements by role.
證據類型 / Evidence type
作者教學。
Author exposition.
限制 / Limits
原作製作方式有推測；相似度是趣味說法，不是實測，也非藝術家全貌。
Original-production details include speculation; similarity is playful, not measured, and does not define the artist.
來源衍生雙語預算 / Source-derived bilingual budget: 139/200; 正文逐字引述 / body quotation: 0.
原創應用（非作者教材）/ Original application (not author teaching material)
練習 / Exercise
原創兩小節：保留骨架，改變密度
Two original bars: Preserve structure, vary density
目的 / Purpose
用原創材料安排角色，不學抄現成作品，也不把機械整齊當作原聲鋼琴的唯一標準。
Arrange roles using original material, without copying an existing work or making mechanical uniformity the sole acoustic-piano standard.
材料 / Materials
紙、筆與下列完整音符表。4/4 在本題表示每小節四拍、每拍一個四分音符；「&」是每拍的後半。
Paper, pencil, and the complete note table below. Here 4/4 means four quarter-note beats per bar; “&” marks the second half of each beat.
完整資料卡 / Complete data card
旋律 / Melody
1.
小節 / Bar
1
拍點 / Beats
1234
音符 / Notes
E4D4G4E4
每音拍數 / Beats per note
1
2.
小節 / Bar
2
拍點 / Beats
1234
音符 / Notes
F4D4E4C4
每音拍數 / Beats per note
1
低音規劃 / Bass plan
低音角色只畫在紙上：第一小節 C3 維持四拍，第二小節 G2 維持四拍；不要求雙手合彈。
Draw the bass role only: C3 for four beats in bar one, G2 for four beats in bar two; two-hand playing is not required.
節奏規劃 / Pulse plan
節奏角色在每小節第 2、4 拍畫點；這些點不是指定任何樂器或音色。
Mark the pulse role on beats 2 and 4 of each bar; dots do not specify an instrument or timbre.
變體規則 / Variant rule
B 版把每個旋律音分成兩個同音的半拍事件；低音與點的位置不變。A 版是原表。
Version B divides each melody note into two repeated half-beat events; bass and dot positions stay unchanged. Version A is the original table.
鍵位說明 / Keyboard help
C4 是中央 C；往右白鍵依序 D4、E4、F4、G4。C3 是下一個較低的 C，G2 是 C3 左方的 G；只讀紙上低音即可。
C4 is middle C; successive white keys to its right are D4, E4, F4, G4. C3 is the next lower C and G2 the G left of C3; reading the bass on paper is sufficient.
步驟 / Steps
先畫兩小節八個拍格，再為每拍分成兩半。寫 A 的旋律、低音與點，避免把三種角色當成三條都要同時彈的作業。 Draw eight beat boxes across two bars, then halve each beat. Enter A’s melody, bass, and dots; these roles are not an obligation to play three parts at once.
先預測 A、B 的旋律事件數與總長度，再依規則寫出 B。問：事件變多是否必然更好聽？ Predict melody-event counts and total duration for A and B, then write B. Ask whether more events necessarily sound better.
挑一個你想改的目標，例如更有空間；只改一個變量，可刪去第二小節第 2 拍的旋律並留一拍休止，其他不變。此改寫不必照藝術家模板。 Choose a goal such as more space and alter one variable: omit the melody on beat 2 of bar two, leaving a one-beat rest and all else unchanged. No artist template is required.
可選單手慢彈 A 的旋律，每音一拍，不踩踏板；B 可以只讀紙上，不做速度挑戰。記錄偏好與理由，不替藝術家或聽眾作客觀評分。 Optionally play A’s melody slowly with one hand, one beat per note, without pedal. B can stay on paper rather than become a speed challenge. Record a preference and reason, not an objective score for the artist or audience.
參考答案 / Answer guidance
A 有 8 個旋律事件，B 有 16 個；兩者仍是 8 拍。B 的第一小節依序 E4/E4、D4/D4、G4/G4、E4/E4；第二小節 F4/F4、D4/D4、E4/E4、C4/C4。 A has 8 melody events and B 16; both last 8 beats. B’s first bar is E4/E4, D4/D4, G4/G4, E4/E4; its second is F4/F4, D4/D4, E4/E4, C4/C4.
增加事件改變密度，不足以證明更好聽或更像任何人。空間版本仍長 8 拍，但 A 的 8 個旋律事件變為 7 個加一拍休止。 More events change density, not prove greater quality or artist similarity. The space variant still lasts 8 beats, with A’s 8 melody events becoming 7 plus a one-beat rest.
完成條件 / Completion check
角色分明、拍數正確、寫出一條有理由的偏好即可；不要求產出電子音色或完成複製作品。
Stop once roles and durations are clear and one preference is justified; no electronic timbre or copied arrangement is required.
回饋規則 / Feedback rule
未來回饋應先問任務是節拍練習或表情演奏；不要把有意伸縮、層次或音色差異全部算錯，也不能從樂器清單推得每個音符。
Future feedback should first ask whether the task is pulse practice or expressive performance. Do not mark all intentional timing, balance, or timbre variation wrong, and do not infer every note from an instrument list.
5. 用自由軟體來「線上練團」好好玩！ / Online ensemble practice with free software
來源 / Source: WEB-e4dc31285955 · 原文 / Primary article
發表 / Published: 2021-06-26 · 閱讀 / Read: 2026-10-22
涵蓋 / Coverage: 30–103 / 106 lines
完整正文；媒體未審。
Complete body; media unreviewed.
來源摘要 / Source summary
追蹤收音、編碼、網路與播放延遲，介紹 Jamulus 合奏案例。
Tracks capture, codec, network, and playback delays through a Jamulus ensemble case.
內容角色 / Content role
分開同步與輪流。
Separate synchrony from turn-taking.
證據類型 / Evidence type
作者教學。
Author exposition.
限制 / Limits
約 20 ms 是作者舒適目標，非普遍門檻；2021 年平台敘述非現況。
About 20 ms is the author’s comfort target, not a universal cutoff; 2021 platform claims are historical.
來源衍生雙語預算 / Source-derived bilingual budget: 123/200; 正文逐字引述 / body quotation: 0.
原創應用（非作者教材）/ Original application (not author teaching material)
練習 / Exercise
延遲帳本：相同網速，不同任務
Latency ledger: Same connection, different tasks
目的 / Purpose
先界定要一起彈還是輪流示範；以紙上算術避免把傳輸延遲全算成學生節拍錯誤。
Define simultaneous playing versus alternating demonstrations first; use paper arithmetic to avoid assigning all transmission delay to learner timing.
材料 / Materials
紙、筆及以下虛構毫秒表；1 ms=0.001 秒，BPM 是每分鐘拍數，本題一拍為四分音符。沒有進行連線測試。
Paper, pencil, and the fictional millisecond table below. 1 ms=0.001 seconds; BPM is beats per minute, with a quarter-note beat here. No connection test occurred.
完整資料卡 / Complete data card
單向音訊路徑（ms）/ One-way audio path (ms)
1.
環節 / Stage
A 輸入
A input
毫秒 / Milliseconds
4
2.
環節 / Stage
A 打包／編碼
A framing/encoding
毫秒 / Milliseconds
3
3.
環節 / Stage
A 到伺服器
A to server
毫秒 / Milliseconds
6
4.
環節 / Stage
伺服器處理
Server processing
毫秒 / Milliseconds
2
5.
環節 / Stage
伺服器到 B
Server to B
毫秒 / Milliseconds
8
6.
環節 / Stage
B 解碼
B decoding
毫秒 / Milliseconds
3
7.
環節 / Stage
B 輸出
B output
毫秒 / Milliseconds
4
範圍 / Scope
這是 A 到 B 的一次音訊路徑；各列定義為互不重疊，未另計房間空氣傳播。不要再把完整總和加一次。
This is one A-to-B audio path; rows are defined as non-overlapping, excluding additional room-air propagation. Do not add the full total twice.
網路探測 / Network probe
另有 A→伺服器→A 的網路往返測值 12 ms。它不是 A→B 的音訊總和。
A separate A→server→A network round trip is 12 ms. It is not the A→B audio total.
時序 / Timing
速度（BPM）/ Tempo (BPM)
120
後續單向音訊延遲（ms）/ Later one-way audio delays (ms)
303555
公式 / Formula
每拍毫秒=60000/BPM；十六分音符時長=每拍毫秒/4；延遲佔拍數=延遲毫秒/每拍毫秒。
Milliseconds per beat=60000/BPM; sixteenth-note duration=milliseconds per beat/4; delay in beats=delay milliseconds/milliseconds per beat.
步驟 / Steps
先預測路徑總和，再逐列相加；把網路往返與音訊單向兩個量寫在不同欄，不可以直接替換。 Predict the path total, then add its rows. Put network round trip and one-way audio in separate columns rather than substituting one for the other.
算每拍與十六分音符長度，再算總和佔多少拍。這只描述時差，不能決定每個人的舒適度。 Calculate beat and sixteenth-note duration, then the total as a fraction of a beat. This describes delay, not everyone’s comfort.
對後續 30、35、55 ms 寫出最小、最大與範圍；延遲變動可稱抖動，本題只算範圍，不將其冒稱標準化網路指標。 For later values 30, 35, 55 ms, find minimum, maximum, and range. Delay variation may be called jitter; this exercise calculates only range, not a standardized network metric.
為「同時合奏」和「學生彈完，老師再說」各寫所需證據。對前者要完整、穩定且適合曲目與參與者的端到端時序證據；本卡並沒有提供實際驗證。 State evidence needed for simultaneous ensemble and for “learner finishes, teacher responds.” The former needs complete, stable end-to-end timing suitable for the music and participants; this card supplies no actual validation.
若不能建立同步條件，改用輪流示範或紙上拍點，不能偷偷升級成買設備、改網路或開公開房間的任務。 If synchrony is unestablished, use alternating demonstrations or paper beat marking, without silently escalating to equipment purchases, network changes, or public rooms.
參考答案 / Answer guidance
總和=4+3+6+2+8+3+4=30 ms；每拍 500 ms，十六分音符 125 ms；30/500=0.06 拍，30/125=0.24 個十六分音符。 Total=4+3+6+2+8+3+4=30 ms; a beat is 500 ms and a sixteenth 125 ms; 30/500=0.06 beats and 30/125=0.24 sixteenths.
後續最小 30、最大 55、範圍 25 ms；不能由平均值或單一 ping 保證合作穩定，也不能將 55 ms 全歸為學生拖拍。 Later minimum is 30, maximum 55, and range 25 ms. An average or single ping cannot guarantee stable coordination; 55 ms cannot all be assigned to learner lateness.
輪流教學不需要兩端音符同時對齊，仍需聽得清楚、輪到誰與結束訊號可辨；它不是任意糟糕網路都能用的保證。 Turn-taking does not require simultaneous note alignment across endpoints, but intelligibility, turn ownership, and ending cues still matter; it is not a guarantee under arbitrarily poor connectivity.
完成條件 / Completion check
完成總和、兩種比例與範圍，並寫出不同任務的條件即可；保留實際連線能力未知。
Finish the total, both ratios, range, and task-specific conditions; leave actual connection capability unknown.
原聲鋼琴界線 / Acoustic-piano boundary
原聲鋼琴會在房間內直接發聲，不能當作可靜音的電子鍵盤；此處不要求遮蔽自然聲、加大耳機或更動任何收音鏈。
An acoustic piano sounds directly in the room and is not a muteable electronic keyboard; this task does not require masking that sound, raising headphone volume, or changing capture routing.
6. 不要再 Zoom 了！用自由軟體來上遠端音樂課吧～ / Remote music lessons with free software rather than Zoom
來源 / Source: WEB-2fccbe9dee7b · 原文 / Primary article
發表 / Published: 2021-05-31 · 閱讀 / Read: 2026-10-22
涵蓋 / Coverage: 32–267 / 270 lines
完整正文；媒體未審。
Complete body; media unreviewed.
來源摘要 / Source summary
介紹自架 Jitsi Meet、登入控制與音樂音訊設定，提醒關閉回聲處理的代價。
Introduces self-hosted Jitsi Meet, login control, and music-audio settings, noting consequences of disabling echo processing.
內容角色 / Content role
設計遠端課流程。
Design remote-lesson workflow.
證據類型 / Evidence type
作者教學。
Author exposition.
限制 / Limits
2021 年步驟、價格與安全比較未驗證現況；自行架設不自動安全，音質不等於同步能力。
2021 steps, prices, and security comparisons are not currently validated; self-hosting is not automatically secure, and fidelity is not synchrony.
來源衍生雙語預算 / Source-derived bilingual budget: 156/200; 正文逐字引述 / body quotation: 0.
原創應用（非作者教材）/ Original application (not author teaching material)
練習 / Exercise
不用開會的遠端課桌上演練
A remote-lesson tabletop rehearsal without a call
目的 / Purpose
在沒有權限、帳號或設備變更的情況下，練習讓老師的問題可回答、學生的表現不被傳輸品質誤判。
Without permissions, accounts, or equipment changes, practice making teacher questions answerable and preventing transmission quality from being mistaken for learner performance.
材料 / Materials
紙、筆與以下三張情境卡。可一人扮演兩個角色；不需真的聯絡老師或進入視訊房間。
Paper, pencil, and the three scenario cards below. One person can play both roles; no teacher contact or meeting room is needed.
完整資料卡 / Complete data card
課程目標 / Lesson target
本次只問四音短句的第二個音是 D4 還是 E4；不評力度、踏板或姿勢。
Ask only whether the second note of a four-note phrase is D4 or E4; do not assess dynamics, pedal, or posture.
預定音符 / Intended notes
C4D4E4C4
每音拍數 / Beats per note
1
情境卡 / Scenario cards
1.
識別 / ID
L1
證據 / Evidence
老師端只收到 C4、未知、E4、C4；學生自述第二音是 D4。
The teacher-side record contains C4, unknown, E4, C4; the learner reports D4 as the second note.
2.
識別 / ID
L2
證據 / Evidence
老師端被告知末尾消失，但沒有原始記錄或設定履歷。
The teacher-side ending is reported missing, with no original record or settings history.
3.
識別 / ID
L3
證據 / Evidence
學生拒絕麥克風；仍願意在紙上寫出預定四音與自己最不確定的位置。
The learner declines microphone access but will write the intended four notes and the location they are least sure about.
輪替流程 / Turn protocol
桌上流程：老師寫一個問題→學生指讀或可選試彈→學生寫「完成」→老師等此訊號後回應；不要兩端一起追著拍點彈。
Tabletop sequence: teacher writes one question → learner points to notes or optionally plays → learner writes “done” → teacher responds after that cue; do not have both ends chase a shared beat.
步驟 / Steps
先寫 L1–L3 各能確定與不能確定的一件事。把目標音、學生自述與教師端證據分開。 For L1–L3, first write one known and one unknown. Separate intended notes, learner report, and teacher-side evidence.
用桌上流程演練一輪；沒有連線也能完成。若選擇試彈，只在自己的原聲鋼琴單手慢彈四音，每音一拍，不踩踏板；不開任何音訊程式。 Rehearse one tabletop round without connecting. If playing is chosen, use one hand slowly on the acoustic piano, one beat per note, no pedal; launch no audio application.
老師回覆只能針對有證據的一點，例如先確認紙上目標，再把聲音判讀標為未知；不可把學生拒絕權限當作演奏零分。 The teacher response addresses one evidenced point, for example confirming the written target while marking sound interpretation unknown. Denying permissions must not become a zero performance score.
列出若未來真的上課才需另外確認的事：同意、聯絡對象、版本與設定、可理解的音訊、誰可進入、是否錄製及保留多久。這是待確認清單，不是現在授權。 List matters to confirm separately before any future real lesson: consent, participants, version/settings, intelligible audio, room access, recording, and retention. This is a pending checklist, not current authorization.
參考答案 / Answer guidance
L1：目標第二音是 D4，但缺失的接收證據不能證明實際彈 D4 或 E4；自述另列。L2：尾音狀態與原因未知，不能診斷放鍵、踏板或觸鍵。 L1: D4 is the target second note, but missing received evidence cannot prove that D4 or E4 was played; keep the self-report separate. L2: ending and cause are unknown; do not diagnose release, pedal, or touch.
L3：可以完成紙上預測與不確定位置標記，不能宣稱已驗證聲音表現。輪流流程容許回應等待，卻不能修復被漏掉的證據。 L3: written prediction and uncertainty marking remain possible, but sound performance is unverified. Turn-taking accommodates response waits; it does not repair missing evidence.
完成條件 / Completion check
完成三張卡的已知／未知、一次桌上輪替與一條不強求權限的回饋即可。此題不證明任何平台適用、安全或教學有效。
Stop after known/unknown entries for three cards, one tabletop exchange, and a permission-respecting response. This does not establish platform suitability, security, or teaching efficacy.
回饋規則 / Feedback rule
產品應允許「無法判讀」與紙上替代路線；登入或房間狀態不能作為音訊準確度證據，也不把配置文件讀取當作實機驗證。
A product should offer “uninterpretable” and a paper alternative. Login or room status is not audio-accuracy evidence, and reading configuration docs is not live verification.
補充一手核查 / Supplementary primary-source checks
SUP17-ALLPASS: 全通濾波器 / Allpass Filters — Julius O. Smith III
一手來源 / Primary source · 20–25; 31 · 2026-10-22
指定文字；未測產品或媒體。
Specified text; no product or media test.
此書的全通模型各頻率振幅增益為一，相位可不同。
In this book’s allpass model, amplitude gain is one at every frequency while phase can differ.
作者技術書的概念，不是效果器實測。
Author technical-book concept, not an effect test.
編輯推論 / Editorial inference
本批推論：與原訊號相加後，頻譜不一定平坦；加法題示範抵消的可能，不是效果器實測。
Our inference: after mixing with the original, the spectrum need not be flat. The addition exercise illustrates possible cancellation, not a tested effect.
雙語預算 / Bilingual budget: 148/200.
SUP17-RECOVERY: RX 8 修復技術介紹 / Reinventing Restoration with RX 8: Wow, Flutter, Guitar De-noise, and Spectral Recovery
一手來源 / Primary source · 628–650 · 2026-10-22
指定文字；未測產品或媒體。
Specified text; no product or media test.
廠商在 2020 年描述頻譜恢復為估計缺失頻率並生成替代內容，不是保證取回原樣本。
The vendor’s 2020 account describes estimating missing frequencies and generating replacement content, not guaranteed retrieval of original samples.
行銷示例與媒體未實測；不推論目前版本效果。
Marketing examples and media untested; no current-version efficacy inferred.
發表 / Published: 2020-09-15
雙語預算 / Bilingual budget: 121/200.
SUP17-JAMULUS: Jamulus 使用手冊 / User Manual — Jamulus
一手來源 / Primary source · 93–114 · 2026-10-22
指定文字；未測產品或媒體。
Specified text; no product or media test.
手冊分開網路 ping 與含音訊緩衝的總延遲；中斷可能涉及緩衝、頻寬或負載。
The manual separates network ping from total delay including audio buffers; interruptions may involve buffers, bandwidth, or load.
只核實文件敘述，不認證任何人的連線。
Document wording only, not certification of any connection.
已觀察官方網址重新導向 / Observed official URL redirect: https://jamulus.io/wiki/Software-Manual → https://jamulus.app/wiki/Software-Manual.
雙語預算 / Bilingual budget: 167/200.
SUP17-JITSI: Jitsi Meet 設定文件 / Configuration — Jitsi Meet
一手來源 / Primary source · 203–214; 279–287 · 2026-10-22
指定文字；未測產品或媒體。
Specified text; no product or media test.
文件稱音訊品質設定會停用回聲消除、降噪與自動增益；吵雜麥克風偵測是警告功能，並非降噪本身。
The document says its audio-quality configuration disables echo cancellation, noise suppression, and automatic gain control; noisy-microphone detection is a warning feature, not noise suppression itself.
未檢查部署版本或實機；不照抄舊設定。
Deployment version and behavior untested; do not copy old settings.
頁面更新 / Page updated: 2026-10-13
雙語預算 / Bilingual budget: 121/200.
跨批預算 / Cross-batch budgets
已搜尋既有 public-batch*.json 的精確補充網址／識別片段；本批四個補充來源未見先前重用。未重新引用第 12、15 批已近上限的 MDN、Apple、Audacity 頁面。未來整合若重述任何來源內容，須按網址合併重新計數；兩個輸出格式僅算一份。
Existing public-batch*.json files were searched for exact supplementary URLs/identifying fragments; none of these four supplements was previously reused. Near-limit MDN, Apple, and Audacity pages from batches 12 and 15 are not recited again. Future restatements must be aggregated by URL and recounted; the two output formats count as one copy.
原創題目只有新定義、資料與推論；不另延伸來源摘要。
Original exercises supply new definitions, data, and reasoning, not extended source summaries.
初學者路線 / Beginner route
先做第六卡：只需分開已知與未知，建立輪流回饋。 Start with card six: separate known and unknown and rehearse turn-taking.
再做第三、二卡：用加法與截頂認識不可逆與不唯一，不必先學聲學公式。 Next use cards three and two: addition and clipping illustrate non-uniqueness without advanced acoustics.
第四卡可做原創編排；第一與第五卡是選修概念與算術。可分日完成，不要求一次全部練完。 Card four provides original arranging; cards one and five are optional conceptual arithmetic. Spread the work across days rather than requiring all cards at once.
原創回饋產品驗收設計 / Original feedback-product acceptance design
狀態 / Status
以下是原創需求與桌上驗收，尚未建立或測試應用。
These are original requirements and tabletop acceptance checks; no app was built or tested.
檢查 / Checks
輸入只有修復或編輯版本時，顯示版本限制，不把修改後音高或拍點當作原演奏分數。 When only restored or edited material is supplied, show version limits and do not score modified notes/timing as the original performance.
對傳輸缺片段，回傳「資料不足」而非錯音、拖拍或踏板錯誤；保留位置與所缺證據。 For transmission gaps, return insufficient evidence rather than wrong-note, lateness, or pedal diagnoses; retain the location and missing evidence.
問清同步合奏或輪流示範；沒有完整端到端量測時，不從文字研究推定實際可同步。 Distinguish simultaneous ensemble from alternating demonstrations. Without full end-to-end measurements, do not infer actual synchrony from text research.
使用者拒絕麥克風後，六卡紙上核心仍完整；不暗中錄音、加入會議、開帳號或更改設定。 After microphone refusal, all six paper tasks remain complete; do not secretly record, join meetings, create accounts, or change settings.
來源摘要與原創作業分欄；英文、中文都能讀到限制與答案，不能只翻題目而漏掉風險。 Separate source summaries from original exercises. Both languages must include limits and answer guidance, not just translated prompts.
測試資料全用本批虛構序列；任何未來真實錄音評估另需授權、來源、比較基準與適用範圍，不用藝術家作品替代。 Use this batch’s fictional sequences for tabletop checks. Future real-recording evaluation separately requires authorization, provenance, references, and scope, not substitution of artist works.
必要紀錄欄位 / Required record fields
題目與目標 Task and target
版本與處理履歷 Version and processing history
證據位置與缺口 Evidence location and gaps
已知／未知／自述 Known/unknown/self-report
預測、答案是否已見 Prediction and answer-exposure state
觀察與主觀偏好分欄 Separate observation and subjective preference
未審媒體與驗證界線 / Unreviewed media and verification boundary
若未來獲准做實聽審查，另記可用媒體、實際聆聽範圍、比較條件與不能判定項；本批不可自動升格為音訊、影片或軟體評測。
If listening review is later authorized, separately log available media, actual listening coverage, comparison conditions, and unresolved questions. This batch must not be automatically upgraded into audio, video, or software evaluation.
用普通本機 Python 核對原創算術與來源字數；沒有模型推理或音訊處理。
Ordinary local Python checks original arithmetic and source budgets; no model inference or audio processing.
六篇正文已讀；來源與算術自檢完成，等待獨立審核。
Six article bodies read; source and arithmetic self-checks complete, pending independent review.