2007:Multiple Fundamental Frequency Estimation & Tracking Results

From MIREX Wiki
Revision as of 10:39, 26 July 2010 by Singh14 (talk | contribs) (For Onset/Offset Evaluation)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Introduction

These are the results for the 2007 running of the Multiple Fundamental Frequency Estimation and Tracking task. For background information about this task set please refer to the 2007:Multiple Fundamental Frequency Estimation & Tracking page.


General Legend

Team ID

CC1 = Chuan Cao, Ming Li, Jian Liu, Yonghong Yan 1
CC2 = Chuan Cao, Ming Li, Jian Liu, Yonghong Yan 2
AC1 = Arshia Cont 1
AC2 = Arshia Cont 2
AC3 = Arshia Cont 3
AC4 = Arshia Cont 4
KE1 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama 1
KE2 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama 2
KE3 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama 3
KE4 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama 4
VE1 = Valentin Emiya, Roland Badeau, Bertrand David 1
VE2 = Valentin Emiya, Roland Badeau, Bertrand David 2
PL = Pierre Leveau
PI1 = Antonio Pertusa, José Manuel Iñesta 1
PI2 = Antonio Pertusa, José Manuel Iñesta 2
PI3 = Antonio Pertusa, José Manuel Iñesta 3
PE1 = Graham Poliner, Daniel P. W. Ellis 1
PE2 = Graham Poliner, Daniel P. W. Ellis 2
SR = Stanisław A. Raczyński, Nobutaka Ono, Shigeki Sagayama
RK = Matt Ryynänen, Anssi Klapuri
EV1 = Emmanuel Vincent, Nancy Bertin, Roland Badeau 1
EV2 = Emmanuel Vincent, Nancy Bertin, Roland Badeau 2
EV3 = Emmanuel Vincent, Nancy Bertin, Roland Badeau 3
EV4 = Emmanuel Vincent, Nancy Bertin, Roland Badeau 4
CY = Chunghsin Yeh
ZR = Ruohua Zhou, Joshua D. Reiss

Overall Summary Results Task1

Below are the average scores across 28 test files. These files consisted of 7 groups, each group having 4 files ranging from 2 polyphony to 5 polyphony. 20 real recordings, 8 synthesized from RWC samples.

AC1 AC2 CC1 CC2 CY EV1 EV2 KE1 KE2 PE1 PI1 PL RK SR VE ZR
Precision 0.298 0.373 0.567 0.359 0.765 0.659 0.687 0.335 0.348 0.734 0.827 0.689 0.690 0.614 0.530 0.710
Recall 0.530 0.431 0.671 0.767 0.655 0.513 0.625 0.618 0.546 0.505 0.608 0.417 0.709 0.595 0.157 0.661
Acc 0.277 0.311 0.510 0.359 0.589 0.466 0.543 0.327 0.336 0.444 0.580 0.394 0.605 0.484 0.145 0.582
Etot 1.444 0.990 0.685 1.678 0.460 0.594 0.538 1.427 1.188 0.639 0.445 0.639 0.474 0.670 0.957 0.498
Esubs 0.332 0.348 0.200 0.232 0.108 0.171 0.135 0.339 0.401 0.120 0.094 0.151 0.158 0.185 0.070 0.141
Emiss 0.138 0.221 0.128 0.001 0.238 0.317 0.240 0.046 0.052 0.375 0.298 0.432 0.133 0.219 0.767 0.197
Efa 0.974 0.421 0.356 1.445 0.115 0.107 0.163 1.042 0.734 0.144 0.053 0.055 0.183 0.265 0.120 0.160

download these results as csv

Where

2007 ev formulas.png

  • Nref is the number of non-zero elements in the ground truth data.
  • Nsys is the number of active elements returned by the system.
  • Ncorr is the number of correctly identified elements.

Individual Results Files for Task 1

Individual results: Scores per Query

AC1 = Arshia Cont
AC2 = Arshia Cont
CC1 = Chuan Cao, Ming Li, Jian Liu, Yonghong Yan 1
CC2 = Chuan Cao, Ming Li, Jian Liu, Yonghong Yan 1
CY = Chunghsin Yeh
EV1 = Emmanuel Vincent, Nancy Bertin, Roland Badeau
EV2 = Emmanuel Vincent, Nancy Bertin, Roland Badeau
KE1 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama
KE2 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama
PE = Graham Poliner, Daniel P. W. Ellis
PI1 = Antonio Pertusa, José Manuel Iñesta 1
PL = Pierre Leveau
RK = Matt Ryynänen, Anssi Klapuri
SR = Stanislaw A. Raczynski, Nobutaka Ono, Shigeki Sagayama
VE1 = Valentin Emiya, Roland Badeau, Bertrand David
ZR = JRuohua Zhou, Joshua D. Reiss

Info about the filenames

The filenames starting with part* comes from acoustic woodwind recording, the ones starting with RWC are synthesized. The legend about the instruments are:

bs = bassoon

cl = clarinet

fl = flute

hn = horn

ob = oboe

vl = violin

cel = cello

gtr = guitar

sax saxophone

bass = electric bass guitar

Run Times

Participant Run time (sec) Machine
CC1 2513 ALE Nodes
CC2 2520 ALE Nodes
KE1 38640 ALE Nodes
KE2 19320 ALE Nodes
VE 364560 ALE Nodes
RK 3540 SANDBOX
CY 132300 ALE Nodes
PL 14700 ALE Nodes
ZR 271 BLACK
SR 41160 ALE Nodes
AP1 364 ALE Nodes
EV1 2366 ALE Nodes
EV2 2233 ALE Nodes
PE1 4564 ALE Nodes
AC1 840 MAC
AC2 840 MAC

download these results as csv

Overall Summary Results Task II

This subtask is evaluated in two different ways. In the first setup , a returned note is assumed correct if its onset is within +-50ms of a ref note and its F0 is within +- quarter tone of the corresponding reference note, ignoring the returned offset values. In the second setup, on top of the above requirements, a correct returned note is required to have an offset value within 20% of the ref notes duration around the ref note`s offset, or within 50ms whichever is larger. The Overlap ratio is calculated for an individual correctly identified note as 2007 overlap.png

A total of 30 files were used in this task: 16 real recordings, 8 synthesized from RWC samples, and 6 piano. The results below are the average of these 30 files.

Results Based on Onset Only

AC3 AC4 EV3 EV4 KE3 KE4 PE2 PI2 PI3 RK VE2
Precision 0.067 0.070 0.412 0.447 0.216 0.263 0.533 0.371 0.203 0.578 0.338
Recall 0.137 0.172 0.554 0.692 0.323 0.301 0.485 0.474 0.296 0.678 0.171
Ave. F-measure 0.087 0.093 0.453 0.527 0.246 0.268 0.485 0.408 0.219 0.614 0.202
Ave. Overlap 0.523 0.536 0.622 0.636 0.610 0.557 0.740 0.665 0.628 0.699 0.486

download these results as csv

Results Based on Onset-Offset

AC3 AC4 EV3 EV4 KE3 KE4 PE2 PI2 PI3 RK VE2
Precision 0.013 0.015 0.142 0.162 0.071 0.063 0.305 0.206 0.094 0.312 0.098
Recall 0.028 0.044 0.226 0.277 0.130 0.084 0.278 0.262 0.130 0.382 0.052
Ave. F-measure 0.020 0.023 0.166 0.204 0.090 0.069 0.277 0.226 0.101 0.337 0.076
Ave. Overlap 0.825 0.831 0.850 0.859 0.847 0.864 0.879 0.844 0.857 0.884 0.804

download these results as csv

Piano Results based on Onset Only

AC3 AC4 EV3 EV4 KE3 KE4 PE2 PI2 PI3 RK VE2
Precision 0.169 0.159 0.591 0.541 0.400 0.441 0.672 0.320 0.240 0.720 0.653
Recall 0.147 0.174 0.651 0.723 0.421 0.361 0.630 0.376 0.194 0.669 0.156
Ave. F-measure 0.168 0.183 0.602 0.601 0.405 0.386 0.647 0.343 0.186 0.692 0.230
Ave. Overlap 0.438 0.428 0.513 0.543 0.440 0.368 0.615 0.495 0.573 0.606 0.516

download these results as csv

Individual Results Files for Task 2

For Onset only Evaluation

AC3 = Arshia Cont
AC4 = Arshia Cont
EV3 = Emmanuel Vincent, Nancy Bertin, Roland Badeau
EV4 = Emmanuel Vincent, Nancy Bertin, Roland Badeau
KE3 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama
KE4 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama
PE2 = Graham Poliner, Daniel P. W. Ellis
PI2 = Antonio Pertusa, José Manuel Iñesta 1
PI3 = Antonio Pertusa, José Manuel Iñesta 1
RK = Matt Ryynänen, Anssi Klapuri
VE2 = Valentin Emiya, Roland Badeau, Bertrand David

For Onset/Offset Evaluation

AC3 = Arshia Cont
AC4 = Arshia Cont
EV3 = Emmanuel Vincent, Nancy Bertin, Roland Badeau
EV4 = Emmanuel Vincent, Nancy Bertin, Roland Badeau
KE3 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama
KE4 = Koji Egashira, Hirokazu Kameoka, Shigeki Sagayama
PE2 = Graham Poliner, Daniel P. W. Ellis
PI2 = Antonio Pertusa, José Manuel Iñesta 1
PI3 = Antonio Pertusa, José Manuel Iñesta 1
RK = Matt Ryynänen, Anssi Klapuri
VE2 = Valentin Emiya, Roland Badeau, Bertrand David

Info About Filenames

The filenames starting with part* comes from acoustic woodwind recording, the ones starting with RWC are synthesized. The piano files are: RA_C030_align.wav,bach_847TESTp.wav,beet_pathetique_3TESTp.wav,mz_333_1TESTp.wav,scn_4TESTp.wav.note, ty_januarTESTp.wav.note

Run Times

Participant Run time (sec) Machine
AC3 900 MAC
AC4 900 MAC
RK 3285 SANDBOX
EV3 2535 ALE Nodes
EV4 2475 ALE Nodes
KE3 4140 ALE Nodes
KE4 20700 ALE Nodes
PE2 4890 ALE Nodes
PI2 165 ALE Nodes
PI3 165 ALE Nodes
VE 390600 ALE Nodes

download these results as csv