characterizaon-of-melodic-mo*fs- in raag-music-- with...
Post on 13-Feb-2018
221 Views
Preview:
TRANSCRIPT
Characteriza*on of Melodic Mo*fs in Raag Music
with Time-‐series Matching
Contributors: Joe Ross
Kaustuv Ganguli Vedhas Pandit
Under guidance of:
Dr. Pree* Rao 1
Outline
• Mo*va*on (Why?)
• Understanding The Challenges (What?)
• Solu*ons (How?) • Performance
– Training – Tes*ng – Configura*on comparisons
• Future Work
2
Mo:va:on
• Raag iden*fica*on • Raag classifica*on
– Thaat system
• What can we pick up as characteris*c of Raag? • Mo*f iden*fica*on • Mo*f classifica*on
4
Raag 101 • What is Raag?
– Literal meaning • Color/hue (relevance to moods, seasons, day-‐*mes)
– Framework comprising of set of Swaras (notes) – Let’s not forget rules
• Sequence-‐level: Aroha (ascent), Avaroha (descent), Pakad (grip), Chalan (gait)
• Swara-‐level: Vadi (speaking), Samvadi(responding) Varjit (exclusions)
– Scale v/s Raag v/s Tune – Classifica*ons (Jaa*, Thaat) – Alankaras (ornamenta*ons)
5
As an example... Raga Characteris:cs Alhaiya Bilawal
Samay (Time) Morning, well before noon (6AM-‐9AM)
Moods Sringar (joy and love), Karuna (pathos,compassion)
Swaras* in use (Tone material) S R G m P D n N
Vadi: D Samvadi: G
Aroha (Ascent) S R GR G P D ND N S’ (m varjit/omi`ed)
Avaroha (Descent) S’ ND P D n D P m G mR S
Classifica*ons Jaa*: Shadav Sampurna Thaat: Bilawal
Pakad (Characteris*c Phrases)
G~ R G /P (GRGP)
D~ n D \P (DnDP)
D \G G m R G P m G
Comments 'n' is used only in the descent, and always in between the two 'D'-‐s as D n D P
6
UNDERSTANDING THE CHALLENGES So, we have an ancient framework for music which is pre`y gripping, but…
7
SOLUTIONS: Summary:
1. Solve ‘unequal dura*ons’ problem first. (Equilength) 2. Shape varia*ons s*ll. If we observe closely, some shapes repeat. (K-‐means) 3. Within group, intra-‐phrase dura*ons vary. (DTW)
12
Original signal Signals made equilength
For varying phrase dura:ons (Inter-‐phrase variability)
0 (Time in sec) 2
14
Dynamic Time Warping (DTW)
Elena Tsiporkova: h`p://www.psb.ugent.be/cbd/papers/gentxwarper/DTWAlgorithm.ppt 18
Global Constraints
Similarity Measures and Dimensionality Reduc*on Techniques for Time Series Data Mining by C. Cassisi, P. Montalto, M. Alio`a, A. Cannata, A. Pulviren*
Sakoe-‐Chiba Band Itakura Parallelogram
23
Threshold DTW (LCB)
Distribu*on of: Distances between corresponding pitch values upon warping Database: DnDP and mnDP phrase pitch signals (Ashwini Bhide + Manjiri Asnare concerts) Indica*ve of: Varia*on/Error in tracking any specific note (origins: expression/post-‐processing) 24
For shape varia:ons (Inter-‐phrase variability)
Contours (as we saw earlier)
Contours individually (for the sake of clarity)
26
K-‐means Shortcomings
• ‘K’ needs to be known before-‐hand • Different ini*al par**ons can result in different clusters i.e. local minima found
• Does not work well with clusters of different size and density
• When used in conjunc*on with DTW-‐based distance measure: – Objec*ve func*on no longer monotonically decreasing, can run into a loop.
– Triangle inequality not sa*sfied (DTW shortcoming), visual representa*on of the data not possible.
29
Database Song ID Artiste Tala Laya Bandish
Tempo (bpm)
Dur. (min)
Phrase Class
DnDP mnDP GRGP
Char. Seq.
AB Ashwini Bhide Tintal Madhya Kavana
Batariyaa 128 8.85 13 2 31 5
MA Manjiri Asanare Tintal Vilambit Dainyaa Kaahaan 33 6.9 12 1 13 6
SS Shruti Sadolikar Tintal Madhya Kavana
Batariyaa 150 4.15 3 0 14 3
ARK Abdul Rashid
Khan Jhaptal Madhya Kahe Ko Garabh 87 11.9 44 0 0 14
DV Dattatreya Velankar Tintal Vilambit
Dainyaa Kaahaan 35 18.3 14 4 4 10
JA Jasraj Ektal Vilambit Dainyaa Kaahaan 13 22.25 19 18 0 29
AK-1 Aslam Khan Jhumra Vilambit Mangta
Hoon Tere 19 8.06 10 0 8 6
AK-2 Aslam Khan Jhaptal Madhya E Ha
Jashoda 112 5.7 7 0 0 3
AC Ajoy
Chakrabarty Jhumra Vilambit Jago Man
Laago 24 30.3 --- 27 0 ---
Total no. of phrases 122 52 70 76
31
Effect of Global Constraints
41
FA rate (Nega*ves:
594)
Hit rate (Total posi*ves: 366)
No Constraints
Learned Sakoe-‐Chiba
Learned constraints
6.06% 36
70.22% 257
71.86% 263
71.04% 260
11.95% 71
89.62% 328
89.34% 327
87.16% 319
18.01% 107
95.63% 350
95.36% 349
92.35% 338
24.07% 143
97.81% 358
97.54% 357
99.18% 363
Effect of Pitch Quan*za*on
42
FA rate (Nega*ves:
594)
Hit rate (Total posi*ves: 366)
Unquan:zed (threshold dtw) Q24 Q12
6.06% 36
70.22% 257
69.67% 255
67.76% 248
11.95% 71
89.62% 328
89.07% 326
86.61% 317
18.01% 107
95.63% 350
95.08% 348
93.99% 344
24.07% 143
97.81% 358
98.36% 360
97.81% 358
Future work
• Larger database • Use of D-‐DTW • Different step pa`erns and slope-‐weights • Different pa`ern recogni*on techniques
• Extend work to Mukhda level
• Mo*f search • Automa*c segmenta*on
43
Raag 101
• Alankars (Ornamenta*ons) a. Meend (glide) b. Kan (grace notes): Sparsh(i), Krintan(i) c. Kampit (vibra*ng) d. Andolan (oscilla*ons) e. Gamaka (shadow notes) f. Khatka/Gitkari (note cluster in gusto) g. Zamzama (addi*on of notes) h. Murki (short subtle taan, less forceful than f, g)
46
Future work (Mukhda-‐level)
• Extend the work to mukhda-‐level
Ka va n Ba ta ri y a G R G G m n D P
G G G m
R
n D P
55
top related