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.
公開資料第十五批：錄音證據與處理混淆
Public batch 15: recording evidence and processing confounds
為原聲鋼琴學習者與未來回饋工具，審閱六篇公開正文；只設計紙上判讀，不採集或評分使用者。 Review six public article bodies for an acoustic-piano learner and a future feedback tool; design paper interpretation only, without recording or scoring the user.
完成：6/6 篇完整正文；音訊／影片／圖片／互動驗證：0。
Completed: 6/6 full article bodies; audio/video/image/interactive reviews: 0.
閱讀與整合規則 / Reading and integration rules
中文原題、來源 ID 與指派網址原樣保留；英文題名是編輯翻譯。兩種檔案是同一研究成果，不能重複計入完成數。
Retain assigned Chinese titles, source IDs, and URLs exactly; English titles are editorial translations. The two files are one deliverable, not two completed reviews.
已讀文章完整正文；播放器、圖說與文字中的聽感描述不算聽過或看過示範。影片、音訊、圖片、互動及實際工具均未驗證。
Full article bodies were read. Players, captions, and written listening impressions do not constitute media review. Videos, audio, images, interactions, and actual tools remain unverified.
作者教材是一手教學說明，不是所有聲學、產品或學習成效主張的一手實驗。補充官方文件只核對指定技術區別，不增加六篇完成數。
The author’s own teaching is primary exposition, not primary experimentation supporting every acoustic, product, or learning claim. Official supplements verify specified distinctions without increasing the six-article count.
來源衍生文字合併計算中文字符與英數詞元，含雙語題名、用途、證據類型、涵蓋範圍、摘要及限制。原創數據、算式推演與方法學判斷分開標示；無正文逐字引用或移植來源練習。
Source-derived budgets combine CJK characters and Latin/alphanumeric tokens, including bilingual titles, roles, evidence type, coverage, summary, and limitations. Original data, calculations, and methodological reasoning are marked separately; no body quotations or copied exercises.
只做紙上題：不開啟麥克風權限、不錄音、不播放、不安裝、不登入、不購買、不推送，也不做大音量、接線、供電或喇叭實驗。
Paper tasks only: no microphone permission, recording, playback, installation, login, purchase, push, or loud-sound, wiring, power, or speaker experiments.
不依處理過的聲音診斷琴技，不為補償錄音效果而要求改變原聲鋼琴觸鍵。所有假想數據與教學設計未經使用者或成效驗證。
Do not diagnose playing from processed audio or ask an acoustic pianist to change touch to compensate for recording effects. All hypothetical data and teaching designs remain unvalidated with users or learning outcomes.
價格、排名、推薦與產品獨家能力不沿用為今日事實；不推薦購買。聽感差異或偏好不當成人人相同的物理測量。
Do not carry prices, rankings, recommendations, or exclusive product-capability claims forward as current facts; no purchases are recommended. Listening impressions and preferences are not universal physical measurements.
原創共用流程 / Original shared workflow
先辨認比較對象：同一次演奏的衍生版本，還是另一次重彈？再找時間、電平、聲道及剪輯紀錄。未知就寫未知。
First identify whether versions derive from the same performance or different takes. Then seek timing, level, channel, and edit records. Preserve unknowns as unknowns.
分四欄寫：觀察到什麼、可能由何種處理造成、還需什麼證據、現在不能下什麼結論。這是證據練習，不是鋼琴測驗。
Use four columns: observation, possible processing explanation, missing evidence, and conclusion not yet justified. This is an evidence exercise, not a piano examination.
原始檔也可能經裝置自動處理；保留原件、處理史、每次版本關係與分析所用版本。
An original file can already include device processing. Preserve the original, processing history, version lineage, and analyzed version.
完成標準是能指出合理替代解釋與不可識別之處，不是猜出老師偏好的聲音。
Completion means identifying plausible alternatives and what cannot be determined, not guessing an instructor’s preferred sound.
官方補充核實 / Official supplemental verification
檢查研究目錄既有 JSON／Markdown 的網址與文件 ID，四個補充來源未見先前摘要；因此本次合計就是跨批合計。未再取用 batch12-audio 的 Apple 壓縮器、MDN 或其他補充正文。此掃描不把 MD／JSON 的同一內容重複加倍。 Existing research JSON/Markdown were checked by URL and document identity; no prior summaries of these four supplements were found, so current totals equal cross-batch totals. Batch12-audio’s Apple compressor, MDN, and other supplemental bodies were not reused. Mirrored MD/JSON content is not double-counted.
SUP15-CURVE — 壓縮器 / Compressor — Audacity Manual
官方來源 / Official source · 審讀行 / Reviewed lines: 15–34
比率與門檻定靜態曲線；補償回升電平。Attack／Release 漸進反應／恢復，非固定延遲。 Ratio and threshold define a static curve; make-up raises level afterward. Attack/release are gradual response/recovery, not fixed delay.
指定文字；未操作。 Specified text; not operated.
只核對術語。 Terminology only.
來源衍生預算 / Source-derived budget: 193/200；包含下列跨段落重述 / including cross-section restatements: sources[1].original_application.data_card.compression_rule, sources[1].limits.
先前摘要比對：無；本次合計即跨批合計。 / Prior summary matches: none; this total is the cross-batch total.
SUP15-NORMALIZE — 正規化 / Normalize — Audacity Manual
官方來源 / Official source · 審讀行 / Reviewed lines: 6–12; 31–54
峰值正規化設定最高振幅；獨立聲道模式可能改變平衡。 Peak normalization sets maximum amplitude; independent-channel mode can alter balance.
指定文字；未操作。 Specified text; not operated.
只核對術語。 Terminology only.
來源衍生預算 / Source-derived budget: 157/200；包含下列跨段落重述 / including cross-section restatements: sources[1].original_application.data_card.normalization_rule, sources[1].original_application.answer_key[1].
先前摘要比對：無；本次合計即跨批合計。 / Prior summary matches: none; this total is the cross-batch total.
SUP15-GLOSSARY — 音訊詞彙 / Glossary — Audacity Manual
官方來源 / Official source · 審讀行 / Reviewed lines: 14–15; 28; 36; 47; 76
振幅是訊號大小；樣本是瞬時值。dB 是相對參考的對數單位；削波截去越界波形。 Amplitude is signal magnitude; a sample is an instantaneous value. Decibels (dB) are logarithmic relative units; clipping cuts off out-of-range waveform portions.
指定文字；未操作。 Specified text; not operated.
只核對術語。 Terminology only.
來源衍生預算 / Source-derived budget: 191/200；包含下列跨段落重述 / including cross-section restatements: sources[1].original_application.data_card.level_units, sources[1].original_application.data_card.clipping_rule.
先前摘要比對：無；本次合計即跨批合計。 / Prior summary matches: none; this total is the cross-batch total.
SUP15-PANMODES — 設定聲道聲像或平衡 / Set channel strip pan or balance positions in Logic Pro — Apple
官方來源 / Official source · 審讀行 / Reviewed lines: 2286–2294
此版本分單聲道 Pan、立體聲 Balance 與 Stereo Pan；Balance 調兩邊相對電平，Stereo Pan 可各別定位左右輸入。置中增益還受聲像規則影響。 This version distinguishes mono Pan, stereo Balance, and Stereo Pan. Balance changes relative channel levels; Stereo Pan positions left/right inputs individually. Pan law also affects center gain.
指定文字；未操作。 Specified text; not operated.
只核對術語。 Terminology only.
來源衍生預算 / Source-derived budget: 153/200；包含下列跨段落重述 / including cross-section restatements: sources[4].limits.
先前摘要比對：無；本次合計即跨批合計。 / Prior summary matches: none; this total is the cross-batch total.
1. 【無業配】如何買你的第一支麥克風？
[Unsponsored] Choosing your first microphone
WEB-34d692d0306c · 原文 / Source
頁面日期 / Page date: 2024-11-20 · 標示篇幅 / Labeled length: 3105 中文字 / Chinese characters.
審讀正文行 / Body lines reviewed: 38–126（擷取總行 / extraction total 145）。
完整正文；媒體未審。 Complete body; media unreviewed.
作者教學。 Author exposition.
來源重點 / Source insight
USB/XLR 是連接類型；動圈／電容是收音原理；指向性是方向敏感程度。先談環境，無單一優勝類型。 USB/XLR are connection types; dynamic/condenser describe sound-to-signal mechanisms; directionality describes directional sensitivity. Environment comes first, with no intrinsically superior type.
用途 / Role
記錄收音條件。 Document capture conditions.
來源限制與編輯保留 / Source limitations and editorial qualifications
類型概括不等於實測；低靈敏度不自動保證低背景比例。 Type generalizations are not measurements; low sensitivity alone does not guarantee a lower background ratio.
來源衍生預算 / Source-derived budget: 145/200（兩語合計 / both languages combined）。
計入欄位 / Counted fields: title_zh, title_en_editorial, coverage.note, source_claim_summary, content_role, evidence_type, limits.
原創紙上任務 / Original paper task
同倍率不等於去噪 Equal scaling is not denoising
用兩個虛構收音通道，檢查數值大小是否能決定背景比例。 Use two fictional capture channels to test whether raw magnitude determines background proportion.
原創假想紙上題；未產生音訊，未操作軟體。 Original hypothetical paper task; no audio generated or software operated.
資料卡 / Data card
定義 / Definitions S 是標記為鋼琴的成分幅度，N 是標記為背景的成分幅度；兩者由題目提供，不需從混合錄音分離。只有任意單位，非聲壓、靈敏度規格或感知音量。 S is the labeled piano-component amplitude and N the labeled background-component amplitude. Both are supplied, not separated from a mixed recording. Units are arbitrary, not sound pressure, sensitivity specifications, or perceived loudness.
資料列 / Rows1.標記 / ID A
鋼琴成分 S / Piano component S 2
背景成分 N / Background component N 0.4
2.標記 / ID B
鋼琴成分 S / Piano component S 10
背景成分 N / Background component N 2
3.標記 / ID C
鋼琴成分 S / Piano component S 2
背景成分 N / Background component N 0.1
假設條件 / Conditions A/B 由同一對虛構成分經不同統一倍率取得，無自雜訊、失真或頻率差異。C 的位置、指向、環境與倍率均未知。 A/B come from the same fictional components under different uniform multipliers, with no self-noise, distortion, or frequency differences. C has unknown placement, directionality, environment, and scaling.
先作答 / Try first
先算 A/B 的 S÷N。把 B 每個成分除以 5，寫出結果；不要只看哪一筆原始數字小。
Compute S/N for A and B. Divide both B components by 5 and write the result; do not choose by raw magnitude alone.
算 C 的比例，然後從「數值可知」與「不能判定」兩欄分類：背景比例不同、某麥克風類型較好、演奏者觸鍵較穩。
Compute C’s ratio, then classify claims as numerically known or undetermined: different background proportion, a superior microphone type, and steadier pianist touch.
寫一張無購物的資料需求卡：若未來已有錄音，需知道是否同次演奏、位置、方向、裝置處理與電平校準；沒有這些時保留不確定。
Write a no-shopping evidence request: for any future existing recording, ask whether the take, placement, orientation, device processing, and level calibration are known; preserve uncertainty when they are not.
答案與推理 / Answer and reasoning
A＝2÷0.4＝5；B＝10÷2＝5。B÷5 後為 S＝2、N＝0.4，兩者相同。
A = 2/0.4 = 5; B = 10/2 = 5. Dividing B by 5 gives S = 2 and N = 0.4, identical to A.
C＝20。此題只能確認 C 的指定成分比例較高，不能決定原因或購買選項，更不能推論琴技。
C = 20. The supplied component ratio is higher, but its cause and any purchase choice remain undetermined; playing skill is not inferred.
此任務的限制 / Task limitations
比例是題目中可分開的成分幅度比，非一般混合錄音可直接讀取的品質分數；真實房間與裝置不符合全部簡化假設。 The ratio uses separately labeled component amplitudes; it is not a quality score directly readable from ordinary mixed audio. Real rooms and devices need not satisfy these simplified assumptions.
已公開的遷移變題 / Disclosed transfer variant
改題：D 的 S＝3、N＝0.6，比例仍為 5。若更小的顯示數字讓你想判它較乾淨，先核對倍率。 Variant: D has S = 3 and N = 0.6, still a ratio of 5. If smaller displayed numbers tempt a cleaner-recording judgment, check scaling first.
完成檢核 / Completion check
說明計算、指出缺證，拒絕以處理結果診斷琴技。 Explain the calculation, identify missing evidence, and reject skill diagnosis from processing results.
2. 好和弦的混音幼幼班 - 壓縮器（Compressor）篇
Mixing basics: compressor
WEB-94ff2a66f24b · 原文 / Source
頁面日期 / Page date: 2021-09-17 · 標示篇幅 / Labeled length: 3557 中文字 / Chinese characters.
審讀正文行 / Body lines reviewed: 28–141（擷取總行 / extraction total 144）。
完整正文；媒體未審。 Complete body; media unreviewed.
作者教學。 Author exposition.
來源重點 / Source insight
以鼓示範門檻、強弱差、補償與反應／恢復；補償後細聲可更顯著。 Drum examples explain threshold, level contrast, make-up gain, and response/recovery; compensation can expose quiet details.
用途 / Role
審查電平證據。 Audit level evidence.
來源限制與編輯保留 / Source limitations and editorial qualifications
鼓的聽感與設定非鋼琴處方；不把反應時間當純延遲。 Drum impressions and settings are not piano prescriptions; response time is not pure delay.
來源衍生預算 / Source-derived budget: 122/200（兩語合計 / both languages combined）。
計入欄位 / Counted fields: title_zh, title_en_editorial, coverage.note, source_claim_summary, content_role, evidence_type, limits.
原創紙上任務 / Original paper task
三種處理，三種不同證據 Three processes, three different kinds of evidence
不把壓縮、峰值正規化與削波叫成同一件事；分辨錄音電平變化和演奏強弱。 Do not conflate compression, peak normalization, and clipping; distinguish recorded level changes from performed dynamics.
原創假想紙上題；未產生音訊，未操作軟體。 Original hypothetical paper task; no audio generated or software operated.
術語核實參照 / Terminology support: SUP15-CURVE, SUP15-NORMALIZE, SUP15-GLOSSARY.
資料卡 / Data card
電平單位 / Level units 本表 dB 共用假想參考；負值為低於參考。電平差不是聽感差。 These dB levels share a hypothetical reference; negative values are below it. Level differences are not perceived-loudness differences.
輸入電平（dB）/ Input levels (dB)[-24, -12, -6]
壓縮規則 / Compression rule 紙上靜態模型：x 為輸入、y 為輸出（dB）。門檻 T＝−18；壓縮比 3:1（R＝3）。x≤T 時 y＝x，否則 y＝T＋(x−T)÷R。補償另加 5；不模擬時間。 Static paper model: x = input, y = output (dB). Threshold T = −18; ratio 3:1 (R = 3). If x ≤ T, y = x; otherwise y = T + (x − T)/R. Add make-up 5 separately; no time response.
正規化規則 / Normalization rule 獨立單聲道分支：整段＋3 dB，峰值 −6 變 −3；不做偏移校正、分段增益或限幅。 Separate mono branch: add +3 dB throughout, moving peak −6 to −3; no offset correction, segmented gain, or limiting.
削波規則 / Clipping rule 獨立樣本題：[−0.9,−0.3,0.3,0.9] 乘 2，再截到 [−1,1]。這是指定截斷，不是假定浮點超界必然削波。 Separate sample task: multiply [−0.9, −0.3, 0.3, 0.9] by 2, then clamp to [−1, 1]. Clamping is stipulated, not inferred from floating-point over-range.
先作答 / Try first
分別填壓縮、壓縮後補償與正規化三列，不能把三條分支串接。比較各列最高減最低。
Fill separate rows for compression, compression plus make-up, and normalization; do not chain the branches. Compare each row’s maximum minus minimum.
在獨立樣本分支算截斷結果，再全體除以 2。比對是否回到原始四個樣本，說明為何音量縮小不等於修復。
In the separate sample branch, compute the clamped result, then divide everything by 2. Check whether the original four samples return and explain why attenuation is not restoration.
若回饋卡寫「強弱差縮小，證明你彈穩了」，改寫為保留處理鏈的描述；不給觸鍵處方。
If a feedback card says “Smaller dynamic contrast proves steadier playing,” rewrite it with the processing context retained; give no touch prescription.
答案與推理 / Answer and reasoning
壓縮：[−24,−16,−14]，差 10 dB；補償後：[−19,−11,−9]，差仍 10。原始差為 −6−(−24)＝18。
Compression: [−24, −16, −14], span 10 dB. After make-up: [−19, −11, −9], still 10. Original span: −6 − (−24) = 18.
正規化：[−21,−9,−3]，差仍 18。相同峰值不保證相同感知音量，這裡也未測聽感。
Normalization: [−21, −9, −3], still a span of 18. Equal peaks do not establish equal perceived loudness; perception is not tested here.
乘 2 為 [−1.8,−0.6,0.6,1.8]；截斷為 [−1,−0.6,0.6,1]；除 2 為 [−0.5,−0.3,0.3,0.5]，兩端不是原值。
After ×2: [−1.8, −0.6, 0.6, 1.8]; after clamping: [−1, −0.6, 0.6, 1]; after ÷2: [−0.5, −0.3, 0.3, 0.5]. The endpoints are not restored.
可用描述：「此衍生版本的指定電平差由 18 變 10 dB；已知有處理，不能由此判演奏更穩。」
Acceptable wording: “The derivative’s specified level span changed from 18 to 10 dB. Processing is known, so steadier playing is not established.”
此任務的限制 / Task limitations
靜態算式不是實際外掛模擬：未建模偵測器、反應、恢復、前看、頻率或感知音量；表格數值不反推按鍵速度、力量或聽力。 This static calculation is not a plug-in simulation: detector, attack, release, lookahead, frequency, and perceived loudness are unmodeled. Values do not recover key velocity, force, or hearing.
已公開的遷移變題 / Disclosed transfer variant
改題：輸入 −15、T＝−18、R＝3，得 −17。若只告訴你輸出 −17、卻不給處理參數，不能唯一還原演奏電平。 Variant: input −15 with T = −18 and R = 3 gives −17. Knowing only output −17 without processing parameters does not uniquely recover the performed level.
完成檢核 / Completion check
說明計算、指出缺證，拒絕以處理結果診斷琴技。 Explain the calculation, identify missing evidence, and reject skill diagnosis from processing results.
3. 好和弦的混音幼幼班 - Delay 篇
Mixing basics: delay
WEB-167152baf3f6 · 原文 / Source
頁面日期 / Page date: 2021-10-01 · 標示篇幅 / Labeled length: 2160 中文字 / Chinese characters.
審讀正文行 / Body lines reviewed: 32–110（擷取總行 / extraction total 118）。
完整正文；媒體未審。 Complete body; media unreviewed.
作者教學。 Author exposition.
來源重點 / Source insight
區分重複與瀰散殘響；回授是再循環，乾濕是原音／效果比例。 Distinguishes repeats from diffuse reverb; feedback recirculates, while dry/wet balances original/effect.
用途 / Role
追查事件來源。 Trace event provenance.
來源限制與編輯保留 / Source limitations and editorial qualifications
風格聽感非定律；百分比不是感知音量測量。 Style impressions are not laws; percentages do not measure perceived loudness.
來源衍生預算 / Source-derived budget: 159/200（兩語合計 / both languages combined）。
計入欄位 / Counted fields: title_zh, title_en_editorial, coverage.note, source_claim_summary, content_role, evidence_type, limits, original_application.data_card.generator.
原創紙上任務 / Original paper task
誰產生了時間線上的事件？ Who produced the timeline events?
分開親手彈出的起音、處理生成的重複、以及系統時間偏移。 Separate performed onsets, processing-generated repeats, and system time offsets.
原創假想紙上題；未產生音訊，未操作軟體。 Original hypothetical paper task; no audio generated or software operated.
資料卡 / Data card
符號 / Notation 起音＝聲音開始的時點；ms＝毫秒，1000 ms＝1 秒。P/Q 是兩個原始事件標記，不是音名。幅度是模型數，不是聽感百分比。 Onset means a sound’s start time; ms means milliseconds, with 1000 ms = 1 second. P/Q label original events, not pitches. Amplitudes are model values, not loudness percentages.
原始事件 / Original events1.事件 / Event P
時間（毫秒）/ Time (ms) 0
幅度 / Amplitude 1
2.事件 / Event Q
時間（毫秒）/ Time (ms) 600
幅度 / Amplitude 1
副本規則 / Copy rule 保留原始事件，另各加副本：晚 300 ms／幅度 0.4；晚 600 ms／幅度 0.16。只列這兩次。 Retain originals and add copies at +300 ms/amplitude 0.4 and +600 ms/amplitude 0.16. List only these two repeats.
偏移分支 / Offset branch 另一獨立版本只把原始兩事件全體晚移 80 ms，不產生副本。 A separate version shifts both original events 80 ms later without producing copies.
先作答 / Try first
把六個事件依時間排序，重合事件仍保留兩個來源標籤。數事件條目與不同時間點，兩個答案可能不同。
Sort the six event records by time, retaining both provenance labels when they coincide. Count records and distinct timestamps separately.
計算 0.4×0.4。若一拍指定為 600 ms，300 ms 是幾拍？只用紙，不播放。
Compute 0.4 × 0.4. If one beat is stipulated as 600 ms, how many beats is 300 ms? Work on paper only.
比較副本版本與整體偏移版本。若只有處理後時間線，能否把所有事件標成多彈或搶拍？
Compare the copied version with the uniformly shifted version. From the processed timeline alone, can every event be labeled an extra played note or timing error?
答案與推理 / Answer and reasoning
順序：P@0；P1@300；P2@600 與 Q@600；Q1@900；Q2@1200。六條來源記錄、五個不同時間點；600 ms 的重合不是第三次按鍵證據。
Order: P@0; P1@300; P2@600 and Q@600; Q1@900; Q2@1200. There are six provenance records but five distinct times. Coincidence at 600 ms is not evidence of a third keystroke.
0.4²＝0.16；300÷600＝1/2 拍。整體偏移版為 80、680 ms，間隔仍為 600 ms，且只有兩個事件。
0.4² = 0.16; 300/600 = 1/2 beat. The offset version has events at 80 and 680 ms, still 600 ms apart and only two events.
副本、實際重彈與錯誤偵測可在稀疏時間表中混淆；需原始版本與處理紀錄才能作進一步區別。
Copies, actual replaying, and detection errors can be confused in a sparse event table; the original version and processing history are needed for further distinctions.
此任務的限制 / Task limitations
此模型不模擬實際延遲器、連續波形或重疊後的聽感。兩條重合標記不能直接當成兩倍音量；不把系統延遲、回授重複與演奏起音誤差混稱。 This model does not simulate a real delay, continuous waveforms, or overlapping perception. Coincident labels do not automatically mean twice the loudness. System latency, feedback repeats, and performed-onset errors are distinct.
已公開的遷移變題 / Disclosed transfer variant
改題把 Q 放到 650 ms：來源 P2@600 與 Q@650 不再重合。可觀察到 50 ms 間隔，仍不能不查來源就判搶拍。 Move Q to 650 ms: P2@600 and Q@650 no longer coincide. A 50 ms interval is observable, but its provenance must be checked before calling it a timing error.
完成檢核 / Completion check
說明計算、指出缺證，拒絕以處理結果診斷琴技。 Explain the calculation, identify missing evidence, and reject skill diagnosis from processing results.
4. 好和弦的混音幼幼班 - 噪音管理篇
Mixing basics: noise management
WEB-f0b5e8a4dca2 · 原文 / Source
頁面日期 / Page date: 2021-10-30 · 標示篇幅 / Labeled length: 3235 中文字 / Chinese characters.
審讀正文行 / Body lines reviewed: 30–137（擷取總行 / extraction total 140）。
完整正文；媒體未審。 Complete body; media unreviewed.
作者教學。 Author exposition.
來源重點 / Source insight
談剪輯、門控與去噪；承認修音與拉伸可有瑕疵。 Covers editing, gating, denoising, and possible repair/stretch artifacts.
用途 / Role
保留修改履歷。 Preserve edit history.
來源限制與編輯保留 / Source limitations and editorial qualifications
贊助歷史案例，非今日能力排名；語音示範不證鋼琴保真。 Sponsored historical case, not current capability ranking; speech demos do not establish piano fidelity.
來源衍生預算 / Source-derived budget: 162/200（兩語合計 / both languages combined）。
計入欄位 / Counted fields: title_zh, title_en_editorial, coverage.note, source_claim_summary, content_role, evidence_type, limits, original_application.data_card.rule.
原創紙上任務 / Original paper task
把安靜誤認成不重要 Mistaking quiet for unimportant
測試一條簡化刪除規則是否會把想保留的琴音一起消掉。 Test whether a simple deletion rule also removes piano sound that should be retained.
原創假想紙上題；未產生音訊，未操作軟體。 Original hypothetical paper task; no audio generated or software operated.
資料卡 / Data card
規則 / Rule 每窗給非負代表幅度 a。紙上規則 G：a<0.05 則輸出 0，否則留 a；非完整去噪。 Each window supplies nonnegative representative amplitude a. Paper rule G: output 0 if a < 0.05, otherwise a; not complete denoising.
資料列 / Rows1.視窗 / Window A
代表幅度 a / Representative amplitude a 0.02
給定真值標籤 / Supplied truth label 只有背景 Background only
2.視窗 / Window B
代表幅度 a / Representative amplitude a 0.08
給定真值標籤 / Supplied truth label 想保留的輕音與背景 Wanted quiet note plus background
3.視窗 / Window C
代表幅度 a / Representative amplitude a 0.04
給定真值標籤 / Supplied truth label 想保留的琴音尾部 Wanted piano tail
4.視窗 / Window D
代表幅度 a / Representative amplitude a 0.06
給定真值標籤 / Supplied truth label 只有突發雜聲 Unwanted intermittent sound only
標籤限制 / Label limitation 標籤由題目全知提供，實際錄音沒有這個保證；每窗長度未指定，不能推算音符時值。 Labels are supplied by the exercise’s oracle, not guaranteed in real recordings. Window durations are unspecified, so note durations cannot be calculated.
先作答 / Try first
套用 G 並標記：正確刪除、錯刪想留、正確保留、仍留雜聲。
Apply G and label correct removal, unwanted removal of desired material, correct retention, and retained noise.
想讓 C 保留而 D 刪除，只能改一個幅度門檻且仍採「低於門檻即刪」時，做得到嗎？用不等式證明。
Can one amplitude threshold, still deleting values below it, retain C while deleting D? Prove the answer using inequalities.
若另一版本把 C 刪掉，又把 B 時間移動到格線，寫出這會污染哪兩類回饋；保留未改版作比較，不執行修改。
If another version deletes C and moves B to a timing grid, identify two types of feedback it contaminates. Retain an unmodified comparison version; do not perform edits.
答案與推理 / Answer and reasoning
輸出 [0,0.08,0,0.06]：A 正確刪除，B 正確保留但背景也在，C 錯刪，D 雜聲仍留。
Output [0, 0.08, 0, 0.06]: A is correctly removed; B retained with its background; C wrongly removed; D’s unwanted sound retained.
留 C 需 T≤0.04；刪 D 需 T>0.06，不能同時滿足。只憑幅度無法實現指定分類。
Retaining C requires T ≤ 0.04; removing D requires T > 0.06. These cannot both hold, so amplitude alone cannot produce the desired classification.
刪 C 會污染尾音／連接的證據；移 B 會污染時間證據。乾淨或對齊的成品不等於原演奏做對了。
Deleting C contaminates tail/connection evidence; shifting B contaminates timing evidence. A clean or aligned product does not prove the original performance was correct.
此任務的限制 / Task limitations
本題不估計真實噪音，不建議刪除鋼琴休止段，也不把鍵盤或踏板機械聲全部視為無價值。其他去噪方法有不同資訊與失敗方式，未在此驗證。 This task does not estimate real noise, recommend cutting piano rests, or deem all key/pedal mechanics worthless. Other denoising methods have different information and failure modes and are unverified here.
已公開的遷移變題 / Disclosed transfer variant
改題門檻 0.03：A 刪除、B/C/D 保留；想留的尾音回來了，但 D 仍在。寫出取捨，不選一個假裝通用完美的門檻。 Set the threshold to 0.03: A is removed while B/C/D remain. The desired tail survives, but D remains. State the trade-off rather than inventing a universally perfect threshold.
完成檢核 / Completion check
說明計算、指出缺證，拒絕以處理結果診斷琴技。 Explain the calculation, identify missing evidence, and reject skill diagnosis from processing results.
5. 好和弦的混音幼幼班 - Pan 篇
Mixing basics: pan
WEB-97fcd7df1a23 · 原文 / Source
頁面日期 / Page date: 2021-10-09 · 標示篇幅 / Labeled length: 1758 中文字 / Chinese characters.
審讀正文行 / Body lines reviewed: 30–89（擷取總行 / extraction total 92）。
完整正文；媒體未審。 Complete body; media unreviewed.
作者教學。 Author exposition.
來源重點 / Source insight
談左右分離與移動；自動左右移動可能令人暈眩。 Discusses lateral separation and movement; automatic lateral movement may cause dizziness.
用途 / Role
區分聲道與音高。 Separate channel from pitch.
來源限制與編輯保留 / Source limitations and editorial qualifications
其立體聲說明是平衡模式，非所有聲像模式；置中偏好不是定律。 Its stereo description is balance mode, not every panning mode; centering preferences are not laws.
來源衍生預算 / Source-derived budget: 169/200（兩語合計 / both languages combined）。
計入欄位 / Counted fields: title_zh, title_en_editorial, coverage.note, source_claim_summary, content_role, evidence_type, limits, original_application.data_card.terms.
原創紙上任務 / Original paper task
左右通道不是高低音 Left/right channels are not high/low notes
追蹤聲道路由，而非以左右音量判斷音域、左右手或技巧。 Trace channel routing rather than inferring register, hands, or skill from left/right level.
原創假想紙上題；未產生音訊，未操作軟體。 Original hypothetical paper task; no audio generated or software operated.
術語核實參照 / Terminology support: SUP15-PANMODES.
資料卡 / Data card
術語 / Terms 聲道是資料流：單聲道一條、立體聲左右兩條；音域是音高範圍。P 是同一虛構音符。 A channel is a stream: mono has one, stereo left/right. Register is pitch range. P is the same fictional note.
輸入 / Input左 / Left音符標籤 P / Note label P 4
右 / Right音符標籤 P / Note label P 1
映射 / Maps1.標記 / ID A
規則 / Rule L 輸出＝L 輸入×1；R 輸出＝R 輸入×0。 L output = L input × 1; R output = R input × 0.
2.標記 / ID B
規則 / Rule 交換左右完整資料：L 輸出＝原 R；R 輸出＝原 L。 Swap complete streams: L output = original R; R output = original L.
3.標記 / ID C
規則 / Rule 另一獨立單聲道 P＝4；分給 L、R 的倍率各為 1/2。 A separate mono P = 4; assign a multiplier of 1/2 to each of L and R.
模型限制 / Model limit 數值是標記振幅的路由帳本，非混音規格或音高估計；不宣稱符合任何外掛的聲像增益規則。 Values are a routing ledger for labeled amplitudes, not mixing specifications or pitch estimates. No plug-in pan law is claimed.
先作答 / Try first
填出三個映射的 L/R 輸出，並指出哪一個真的把原本右側資料搬到左側。
Compute L/R outputs for all three maps and identify which actually moves originally right-side content to the left.
各版本的音符標籤 P 有沒有改成另一個音高？如果改變左右顯示就判左右手強弱，有什麼缺證？
Has note label P changed into a different pitch? What evidence is missing if a changed left/right display is used to judge hand balance?
把「右邊較小聲，所以高音沒彈好」改成只談可觀察資料的句子。
Rewrite “The right side is quieter, so the high notes were played badly” using only observable information.
答案與推理 / Answer and reasoning
A＝(4,0)，沒有搬移原 R；B＝(1,4)，交換了來源；C＝(2,2)，兩邊含同一 P。
A = (4, 0), without moving original R; B = (1, 4), with swapped sources; C = (2, 2), with the same P in both.
P 未變，題目沒有任何音高變換；也未提供手部、音域、麥克風擺法或未處理音符資料。
P is unchanged; no pitch transformation is specified. Hand, register, microphone arrangement, and unprocessed note evidence are absent.
可寫：「A 的右輸出為零，因規則把它乘零；不能據此判高音或右手。」
Acceptable: “A’s right output is zero because the rule multiplied it by zero; this does not establish high-register or right-hand performance.”
此任務的限制 / Task limitations
同名旋鈕可採不同映射；此題不推測實際軟體。左右相等不保證每位聽者都感到正中，也不保證耳機、喇叭與單聲道回放等效。 Similarly named controls can use different maps; this task does not infer actual software behavior. Equal channels do not guarantee a centered percept for every listener or equivalent headphone, speaker, and mono playback.
已公開的遷移變題 / Disclosed transfer variant
改題讓輸入 L＝1、R＝4，再做 A：輸出 (1,0)。它不是把兩邊混到左邊的 (5,0)；「平衡」與「合併」不能混稱。 Reverse the inputs to L = 1 and R = 4, then apply A: output (1, 0), not the (5, 0) produced by summing both to the left. Balance and summing are distinct.
完成檢核 / Completion check
說明計算、指出缺證，拒絕以處理結果診斷琴技。 Explain the calculation, identify missing evidence, and reject skill diagnosis from processing results.
6. 好和弦的混音幼幼班 - Reverb 篇
Mixing basics: reverb
WEB-440bfa96e870 · 原文 / Source
頁面日期 / Page date: 2021-09-24 · 標示篇幅 / Labeled length: 2768 中文字 / Chinese characters.
審讀正文行 / Body lines reviewed: 30–129（擷取總行 / extraction total 132）。
完整正文；媒體未審。 Complete body; media unreviewed.
作者教學。 Author exposition.
來源重點 / Source insight
談長度、電平及前延遲；快曲可用輕長尾，無唯一正解。 Discusses duration, level, pre-delay; fast music allows quiet long tails; no uniquely correct setting.
用途 / Role
區分尾音與動作。 Separate tails from actions.
來源限制與編輯保留 / Source limitations and editorial qualifications
遠近與清晰度為情境聽感，不是距離或踏板量測。 Distance and clarity impressions are contextual, not distance or pedal measurements.
來源衍生預算 / Source-derived budget: 170/200（兩語合計 / both languages combined）。
計入欄位 / Counted fields: title_zh, title_en_editorial, coverage.note, source_claim_summary, content_role, evidence_type, limits, original_application.data_card.terms.
原創紙上任務 / Original paper task
尾音何時結束，不等於手何時離鍵 A tail’s end is not the hand’s release time
在互不混淆的時間線上放琴鍵、踏板、原始聲音和額外空間響應。 Place key, pedal, source sound, and added spatial response on separate timelines.
原創假想紙上題；未產生音訊，未操作軟體。 Original hypothetical paper task; no audio generated or software operated.
資料卡 / Data card
術語 / Terms 前延遲＝原音與新增響應起點間隔；尾音＝衰退部分。此處乾聲指未加響應，非無房間反射。 Pre-delay is the source-to-added-response onset gap; a tail is decay. Dry here means no added response, not a reflection-free room.
已知動作標籤 / Known action labels按鍵時間（毫秒）/ Key-down time (ms) 0
放鍵時間（毫秒）/ Key-release time (ms) 220
踏板按下 / Pedal down 否 / No
分支 / Branches1.標記 / ID U
原音起點（毫秒）/ Source onset (ms) 0
原音最後越界時點（毫秒）/ Source last above-boundary time (ms) 360
新增響應起點（毫秒）/ Added-response onset (ms) 無新增項目 / No added item
最後越界時點（毫秒）/ Last above-boundary time (ms) 360
2.標記 / ID V
原音起點（毫秒）/ Source onset (ms) 0
原音最後越界時點（毫秒）/ Source last above-boundary time (ms) 360
新增響應起點（毫秒）/ Added-response onset (ms) 45
最後越界時點（毫秒）/ Last above-boundary time (ms) 1100
3.標記 / ID W
原音起點（毫秒）/ Source onset (ms) 0
原音最後越界時點（毫秒）/ Source last above-boundary time (ms) 360
新增響應起點（毫秒）/ Added-response onset (ms) 0
最後越界時點（毫秒）/ Last above-boundary time (ms) 1100
資料卡假設 / Card assumptions 三版本同一假想演奏且全程未踩踏板；動作標籤由外部真值卡提供，不是由聲音估計。聲音超過指定檢測界線的最後時點由題目給出，不是完全無聲的時點；未提供波形。 All versions derive from one hypothetical performance with no pedal use throughout. Action labels come from an external truth card, not audio inference. The last time above a stipulated detection boundary is supplied, not the time of absolute silence; no waveform is given.
先作答 / Try first
把動作與聲音分成四條線。算 V 的新增響應間隔，以及 V 相對 U 最後可檢測點的差。
Draw separate action and sound timelines. Compute V’s added-response onset gap and its last-detectable-time difference from U.
V 和 W 哪些數字不同、哪些相同？只根據表格，能否給實際房間尺寸、距離或偏好排序？
Which numbers differ between V and W, and which match? Can this table alone establish actual room size, distance, or a preference ranking?
現在蓋住動作真值卡。只看 V 的 1100 ms 結尾，可否宣稱踏板踩到 1100 ms 或鍵按到那時？把回饋寫成可辯護句子。
Now hide the action truth card. From V’s 1100 ms endpoint alone, can you claim pedal or key hold until that time? Write a defensible feedback sentence.
答案與推理 / Answer and reasoning
V 新增響應間隔＝45−0＝45 ms；尾部檢測端點差＝1100−360＝740 ms。與放鍵 220 ms 不同，也沒有新增踏板事件。
V’s added-response onset gap is 45 − 0 = 45 ms; the tail-detection endpoint difference is 1100 − 360 = 740 ms. Neither equals key release at 220 ms, and no pedal event was added.
V/W 的新增響應起點不同，最後檢測時點相同；單憑兩個端點不能推房間幾何或人人一致的清晰度。
V/W differ in added-response onset but share the last detected time. Two endpoints alone do not determine room geometry or universally perceived clarity.
可寫：「V 在此檢測規則下仍有尾部到 1100 ms；來源動作尚不能由此恢復。若有真值卡，應另列 220 ms 放鍵、未踩踏板。」
Acceptable: “V retains a tail to 1100 ms under this detection rule; source actions cannot be recovered from that alone. If the truth card is available, separately report key release at 220 ms and no pedal use.”
此任務的限制 / Task limitations
這不是原聲鋼琴衰減模擬；真琴共鳴、鍵與踏板、房間及錄音鏈可能共同影響錄到的尾部。不能為消掉處理尾巴要求更早放鍵或更短踏板；需獨立演奏證據。 This is not an acoustic-piano decay simulation; piano resonance, keys and pedal, room, and recording chain can jointly affect captured tails. Do not prescribe earlier key release or shorter pedaling to remove a processing tail; independent performance evidence is required.
已公開的遷移變題 / Disclosed transfer variant
改題把最後檢測界線提高，端點可能提前，但動作卡未改。答題需區分「檢測規則變了」與「演奏變了」，不猜新的毫秒數。 Raise the detection boundary: the endpoint may move earlier while the action card remains unchanged. Distinguish a changed detection rule from changed playing, without inventing a new millisecond value.
完成檢核 / Completion check
說明計算、指出缺證，拒絕以處理結果診斷琴技。 Explain the calculation, identify missing evidence, and reject skill diagnosis from processing results.
未來回饋工具：原創驗收草案 / Future feedback tool: original acceptance draft
原創驗收草案；沒有建置、執行或宣稱模型通過。 Original acceptance-design draft; no app was built or run, and no model is claimed to pass.
所需資料 / Required metadata
版本：原件 ID、父版本 ID、是否另一次演奏；時間單位與時基。
Version: original ID, parent version ID, retake status, time units, and time base.
收音：裝置、位置、方向、房間條件與已知自動處理；不知道就保留 unknown，不填猜測。
Capture: device, placement, orientation, room conditions, and known automatic processing; retain unknown rather than guessing.
處理：操作順序、電平變換、時間變換、聲道映射、降噪／門控、效果與分析版本；只記有證據的值。
Processing: operation order, level transforms, time transforms, channel maps, denoising/gating, effects, and analyzed version; record only evidenced values.
證據類型：觀察訊號、演算法推測、外部動作真值、人的偏好分欄；信心需有驗證依據，不捏造百分比。
Evidence types: observed signal, algorithmic inference, external action truth, and human preference in separate fields; confidence needs validation rather than invented percentages.
驗收案例 / Acceptance cases
R15-01
只有 A/B 幅度表。 Only the A/B amplitude table.
顯示同一成分比；拒絕選麥克風或診斷觸鍵。 Show equal component ratios; refuse microphone selection or touch diagnosis.
R15-02
同演奏的壓縮版較小強弱差。 Compressed derivative has reduced level contrast.
歸因待處理鏈核查；不得自動加演奏穩定度分數。 Require processing-history attribution; do not automatically raise a steadiness score.
R15-03
原始與副本事件在 600 ms 重合。 Original and copied events coincide at 600 ms.
保留多個來源標籤，不硬算多彈一次；只有輸出時應保留歧義。 Keep multiple provenance labels, not an automatic extra-note penalty; preserve ambiguity when only output exists.
R15-04
門控把指定想留尾音刪除。 Gating deletes a stipulated wanted tail.
警示處理改變證據，不宣稱原演奏斷奏或休止正確。 Flag altered evidence, not original staccato playing or a correct rest.
R15-05
平衡操作把右輸出乘零。 Balance operation multiplies right output by zero.
顯示聲道映射，不推論右手或高音漏彈。 Show channel mapping, not a right-hand or high-note omission.
R15-06
尾部檢測端點 1100 ms，沒有動作真值。 Tail-detection endpoint 1100 ms, without action truth.
不輸出放鍵／踏板精確時間或矯正處方。 Do not output exact key/pedal release times or corrective prescriptions.
R15-07
使用者只想完成紙上知識題。 User only wants a paper knowledge task.
不要求麥克風、上傳、登入、購買或音訊測試；兩語限制同時顯示。 Require no microphone, upload, login, purchase, or audio test; show limitations in both languages.
若任一案例只因數值被後製而判琴技、把缺失資料填成觀察、或只用一種語言呈現重要限制，則不通過。僅有算術測試不能證明實際錄音分析有效。 Fail any case that judges skill from post-processing, fills missing data as observations, or presents important limits in only one language. Arithmetic tests alone do not validate actual recording analysis.
初學順序 / Beginner route
先做麥克風倍率與門控兩題，再做聲道與延遲，最後做壓縮與殘響。可分次完成，無需設備。
Start with capture scaling and gating, then channels and delay, and finally compression and reverb. Split across sessions if useful; no equipment is required.
每題先寫預測再看答案；答錯時檢查單位、分支與來源標籤，不把錯算當成耳朵或琴技不足。
Predict before reading the key. For errors, check units, branches, and provenance labels, not hearing or piano ability.
答案與變題已公開，且未施測；不能視為未見題的遷移測試或成效證據。
Answers and variants are disclosed and unadministered; they are not unseen-transfer tests or evidence of learning outcomes.
驗證狀態 / Verification status
Python 分數算術檢查上述紙上數值、ID／網址／原題一致與字數。不是音訊、軟體或使用者測試。 Python rational-arithmetic checks of paper values, identity preservation, and budget counts; not audio, software-product, or user tests.
算術與資料一致性：通過。 / Arithmetic and data consistency: passed.
日後若另獲授權檢視媒體，需另外記錄片段、實際聽看內容與限制；此次文字審閱不能替代該工作。 Any later authorized media review needs its own excerpt, observed-content, and limitation records; this text review does not substitute for it.
本次阻礙：無。 / Blockers: none.