Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Fundamentals of Music Processing (Chapter 5)
Search
Koga Kobayashi
December 12, 2019
Research
0
86
Fundamentals of Music Processing (Chapter 5)
Koga Kobayashi
December 12, 2019
Tweet
Share
More Decks by Koga Kobayashi
See All by Koga Kobayashi
第13回 Data-Centric AI勉強会, LLMのファインチューニングデータ
kajyuuen
4
1.6k
基礎数学の公式
kajyuuen
1
150
初等確率論の基礎
kajyuuen
1
170
Deep Markov Model を数式で追う (+ Pyroでの追試)
kajyuuen
0
900
完全なアノテーションが得られない状況下での固有表現抽出
kajyuuen
3
3.5k
SecHack365 北海道会 LT
kajyuuen
0
510
専門用語抽出手法の研究と 抽出アプリケーションの開発
kajyuuen
1
1.3k
Other Decks in Research
See All in Research
[輪講] SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features
nk35jk
2
900
Hiding What from Whom? A Critical Review of the History of Programming languages for Music
tomoyanonymous
0
140
Vision and LanguageからのEmbodied AIとAI for Science
yushiku
PRO
1
510
能動適応的実験計画
masakat0
2
790
Delta Airlines® Customer Care in the U.S.: How to Reach Them Now
bookingcomcustomersupportusa
0
110
CSP: Self-Supervised Contrastive Spatial Pre-Training for Geospatial-Visual Representations
satai
3
250
VectorLLM: Human-like Extraction of Structured Building Contours via Multimodal LLMs
satai
4
140
SkySense : A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery
satai
3
310
snlp2025_prevent_llm_spikes
takase
0
140
EOGS: Gaussian Splatting for Efficient Satellite Image Photogrammetry
satai
4
450
20250605_新交通システム推進議連_熊本都市圏「車1割削減、渋滞半減、公共交通2倍」から考える地方都市交通政策
trafficbrain
0
720
業界横断 副業・兼業者の実態調査
fkske
0
230
Featured
See All Featured
Mobile First: as difficult as doing things right
swwweet
224
9.9k
Build The Right Thing And Hit Your Dates
maggiecrowley
37
2.8k
Responsive Adventures: Dirty Tricks From The Dark Corners of Front-End
smashingmag
252
21k
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
30
9.6k
Large-scale JavaScript Application Architecture
addyosmani
512
110k
The Language of Interfaces
destraynor
160
25k
Fashionably flexible responsive web design (full day workshop)
malarkey
407
66k
The Myth of the Modular Monolith - Day 2 Keynote - Rails World 2024
eileencodes
26
3k
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
32
1.5k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
48
9.7k
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
53
2.9k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
27k
Transcript
Fundamentals of Music Processing Chapter 5: Chord Recognition খྛ
ᕣՏ εϥΠυʹؚ·ΕΔਤFundamentals of Music ProcessingΑΓҾ༻
Chapter 5: Chord Recognition Chord(Ի) • 3ͭҎ্ͷҟͳΔԻූ͔Βߏ͞ΕΔԻͷ͜ͱ Harmony() • ෳͷԻ͔ΒͳΔܥྻɺԻਐߦ
FM7 G7 Em7 Am Harmony
Chapter 5: Chord Recognition Chord Recognition(Իೝࣝ) • Ի͔ΒԻਐߦΛೝࣝ͢Δٕज़ ԻָϑΝΠϧ͔ΒίʔυේΛࣗಈͰ࡞ग़དྷΔ Իೝ͕ࣝ͏·͍͘͘ͱ…
Chapter 5.3: HMM-Based Chord Recognition Chapter 5.1~5.2 • ಛྔ͔ΒԻΛ͋Δఔਪఆग़དྷΔ ͔͠͠ɺ͜ΕԻҰͭҰ͔ͭ͠ݟ͍ͯͳ͍
Α͘ग़ΔԻܨ͕Γͷڧ͍Իʹண͍ͨ͠ ྫ: I–IV–V–Iਐߦ • FGසग़͠ɺ͍͖ͳΓFmʹߦ͘͜ͱ΄΅ແ͍ HMM(ӅΕϚϧίϑϞσϧ)Λར༻ͯ͠ԻਪఆΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition Ի: , ঢ়ଶ: (for )
ͱͨ͠ͱ͖ɺϚϧίϑੑΛԾఆ͢Δͱ Ͱ͋Δ֬ A := {α1 , α2 , ⋯, αI } s i ∈ [1 : I] sn+1 = αj P[sn+1 = αj |sn = αi , sn−1 = αk , ⋯] = P[sn+1 = αj |P[sn+1 = αj |sn ] ͜͜Ͱ ΛҎԼͷΑ͏ʹఆٛ͢Δɻ(for ) aij i, j ∈ [1 : I] ͜Εঢ়ଶ͕ ͔Β ʹભҠ͢Δ֬ͱߟ͑Δ͜ͱ͕ग़དྷΔ αi αj aij := P[sn+1 = αj |sn = αi ] ∈ [0,1] ·ͣɺϚϧίϑ࿈Λར༻ͨ͠Ի༧ଌʹ͍ͭͯઆ໌͢Δ
Chapter 5.3: HMM-Based Chord Recognition ۩ମྫ; , ͷͱ͖ I =
3 A := {α1 = C, α2 = G, α3 = F} ·ͨ࠷ॳͷঢ়ଶ͕ Ͱ͋Δ֬ΛҎԼͷΑ͏ʹఆٛ͢Δ αi ci := P[s1 = αi ] ∈ [0,1]
Chapter 5.3: HMM-Based Chord Recognition , , ͷͱ͖ ঢ়ଶܥྻ: ʹ͍ͭͯߟ͑Δ
I = 3 A := {α1 = C, α2 = G, α3 = F} C = (c1 , c2 , c3 )T = (0.6,0.2,0.3)T S = (C, C, C, G, G, F, F, C, C) ࠷ॳͷঢ়ଶ͕ Ͱ͋Δ֬ΛҎԼͷΑ͏ʹఆٛ͠ɺ αi ci := P[s1 = αi ] ∈ [0,1] ۩ମྫ
Chapter 5.3: HMM-Based Chord Recognition ભҠ͕֬ҎԼͷͱ͖ ͷΑ͏ͳԻਐߦ͕ى͖Δ֬ S S =
(C, C, C, G, G, F, F, C, C) = c1 ⋅ a11 ⋅ a11 ⋅ a12 ⋅ a22 ⋅ a23 ⋅ a33 ⋅ a31 ⋅ a11 ≈ 1.29 ⋅ 10−4
Chapter 5.3: HMM-Based Chord Recognition ઌఔঢ়ଶܥྻ Λ༻͍ͯɺԻਐߦͷ֬Λܭࢉ͕ͨ͠ ࣮ੈքͰऔΓ͏Δͯ͢ͷ ʹ͍ͭͯܭࢉෆՄೳ S
S ྫ: 10छྨͷԻɺ20͔ͭΒߏ͞ΕΔۂͷ߹ ύλʔϯͷ֬Λܭࢉ͢Δඞཁ͕͋Δ 1020 ͦ͜Ͱ෦ͷঢ়ଶͰͳ͘ɺಛϕΫτϧΛ༻͍ͯ ԻਐߦΛٻΊΔํ๏ͱͯ͠HMMΛར༻͢Δɻ
Chapter 5.3: HMM-Based Chord Recognition
Chapter 5.3: HMM-Based Chord Recognition
Chapter 5.3: HMM-Based Chord Recognition Input: Իσʔλ͔Β؍ଌͨ͠ܥྻ ͔Β ϞσϧΛར༻͠ɺ؍ଌܥྻ Λ࡞ɻ
O = (o1 , ⋯, oN ) B = (β1 , ⋯, βN ) ؍ଌܥྻ ؍ଌγϯϘϧ ͔Βߏ͞ΕΔ B = (β1 , ⋯, βN ) ℬ = {β1 , ⋯, βk } (for k ∈ [1 : K])
Chapter 5.3: HMM-Based Chord Recognition
Chapter 5.3: HMM-Based Chord Recognition Viterbi: ؍ଌܥྻ ͱ ॳظঢ়ଶͷ֬
ੜ֬ͱભҠ͔֬Β༗άϥϑΛ࡞ B = (β1 , ⋯, βN ) C = (c1 , c2 , c3 )T = (0.6,0.2,0.3)T ੜ֬ ભҠ֬
Chapter 5.3: HMM-Based Chord Recognition ੜ֬ ભҠ֬ ॳظঢ়ଶͷ֬ ੜ͞Εͨ༗άϥϑ
Chapter 5.3: HMM-Based Chord Recognition ViterbiΞϧΰϦζϜʹΑͬͯ ࠷Β͍͠ܦ࿏ʹ͍ͭͯܭࢉΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition ੜ͞Εͨ༗άϥϑ ViterbiΞϧΰϦζϜͰ֤ҐஔͰɺ ֤ԻʹͨͲΓண͘·Ͱͷ࠷దίετͱ ͦͷલʹࢸΔ·ͰͷϙΠϯλΛ֮͑ͳ͕ΒܭࢉΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition ੜ͞Εͨ༗άϥϑ ViterbiΞϧΰϦζϜͰ֤ҐஔͰɺ ֤ԻʹͨͲΓண͘·Ͱͷ࠷దίετͱ ͦͷલʹࢸΔ·ͰͷϙΠϯλΛ֮͑ͳ͕ΒܭࢉΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition ੜ͞Εͨ༗άϥϑ ViterbiΞϧΰϦζϜͰ֤ҐஔͰɺ ֤ԻʹͨͲΓண͘·Ͱͷ࠷దίετͱ ͦͷલʹࢸΔ·ͰͷϙΠϯλΛ֮͑ͳ͕ΒܭࢉΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition ੜ͞Εͨ༗άϥϑ ViterbiΞϧΰϦζϜͰ֤ҐஔͰɺ ֤ԻʹͨͲΓண͘·Ͱͷ࠷దίετͱ ͦͷલʹࢸΔ·ͰͷϙΠϯλΛ֮͑ͳ͕ΒܭࢉΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition ੜ͞Εͨ༗άϥϑ ViterbiΞϧΰϦζϜͰ֤ҐஔͰɺ ֤ԻʹͨͲΓண͘·Ͱͷ࠷దίετͱ ͦͷલʹࢸΔ·ͰͷϙΠϯλΛ֮͑ͳ͕ΒܭࢉΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition ੜ͞Εͨ༗άϥϑ ViterbiΞϧΰϦζϜͰ֤ҐஔͰɺ ֤ԻʹͨͲΓண͘·Ͱͷ࠷దίετͱ ͦͷલʹࢸΔ·ͰͷϙΠϯλΛ֮͑ͳ͕ΒܭࢉΛߦ͏
Chapter 5.3: HMM-Based Chord Recognition ࠷ޙʹɺͦͷϙΠϯλΛḷΓ࠷Β͍͠ԻਐߦΛ ٻΊΔ A := {α1
= C, α2 = G, α3 = F} ͷͱ͖ ̂ S = (C, C, C, G, G, F)