2006:QBSH Discussion Page - Revision history

Kahyun Choi: /* Introduction */

2011-12-20T04:41:45Z

‎Introduction

Mertbay at 18:51, 7 September 2006

2006-09-07T18:51:16Z

Mertbay at 15:35, 7 September 2006

2006-09-07T15:35:17Z

Mertbay at 15:31, 7 September 2006

2006-09-07T15:31:27Z

Jang at 12:45, 7 September 2006

2006-09-07T12:45:11Z

Jang at 11:55, 7 September 2006

2006-09-07T11:55:14Z

Jang at 07:25, 7 September 2006

2006-09-07T07:25:17Z

Rtypke at 15:48, 6 September 2006

2006-09-06T15:48:08Z

Mertbay at 02:26, 6 September 2006

2006-09-06T02:26:11Z

Jang at 16:05, 5 September 2006

2006-09-05T16:05:18Z

@@ Line 1: / Line 1: @@
 =Introduction=
-This page is for discussion concerning the details of the QBSH: Query-by-Singing/Humming task. The final outcomes of our discussions will be posted to the [[QBSH:_Query-by-Singing/Humming]] page.
+This page is for discussion concerning the details of the QBSH: Query-by-Singing/Humming task. The final outcomes of our discussions will be posted to the [[2006:QBSH:_Query-by-Singing/Humming]] page.
 ==Interested Participants ==

← Older revision		Revision as of 18:51, 7 September 2006
Line 246:		Line 246:
	==Mert Bay (20060907)==		==Mert Bay (20060907)==
	One more correction: There will be one flat directory for queries and one for test database, instead of 3. Each song will exist as 3 different files having the same unique id but distinguished by 3 different extensions: .mid,.wav,.pv so that one can choose from each 3 formats. Please make sure your code reads the files appropriately according to their extensions.		One more correction: There will be one flat directory for queries and one for test database, instead of 3. Each song will exist as 3 different files having the same unique id but distinguished by 3 different extensions: .mid,.wav,.pv so that one can choose from each 3 formats. Please make sure your code reads the files appropriately according to their extensions.
		+
		+	==Mert Bay (20060907)==
		+	IMPORTANT:For people who have already submitted or are going to submit: please reread this discussion page, make sure your programs meet necesseary criterions updated as of today. If you have already submitted your code, please go to the submission page:
		+	https://www.music-ir.org/evaluation/MIREX/submission/ , use your existing account(no need to create a new one) and update your package if necessary. Your old package will be overwritten. Please explain how to set the parameters of your program in your readme file.
		+	A summary important changes:
		+	- each program should return 20 candidates per query.
		+	- The test and query databases will be in two different directories where for each song there will be 3 versions, with the same name distinguished by 3 different extensions .mid, .wav , .pv .

← Older revision		Revision as of 15:35, 7 September 2006
Line 245:		Line 245:

	==Mert Bay (20060907)==		==Mert Bay (20060907)==
−	One more correction: There will be one flat directory for queries and one for test database, instead of 3. Each song will exist as 3 different files having the same unique id but ~~with~~ 3 different extensions: .mid,.wav,.pv so that one can choose from each 3 formats. Please make sure your code reads the files appropriately according to their extensions.	+	One more correction: There will be one flat directory for queries and one for test database, instead of 3. Each song will exist as 3 different files having the same unique id but distinguished by 3 different extensions: .mid,.wav,.pv so that one can choose from each 3 formats. Please make sure your code reads the files appropriately according to their extensions.

← Older revision		Revision as of 15:31, 7 September 2006
Line 242:		Line 242:
	==Roger Jang (20060907)==		==Roger Jang (20060907)==
	One more correction: The submitted code should return top-20 candidates instead of 10. The evaluation of both task 1 and 2 will be based on top-20 candidates.		One more correction: The submitted code should return top-20 candidates instead of 10. The evaluation of both task 1 and 2 will be based on top-20 candidates.
		+
		+
		+	==Mert Bay (20060907)==
		+	One more correction: There will be one flat directory for queries and one for test database, instead of 3. Each song will exist as 3 different files having the same unique id but with 3 different extensions: .mid,.wav,.pv so that one can choose from each 3 formats. Please make sure your code reads the files appropriately according to their extensions.

@@ Line 146: / Line 146: @@
 | ALU             |     |   X  |     |    X  |               |
 --------------------------------------------------------------
-| Roger Jang      | X   |      |     |    X  |               |
+| Roger Jang      | X   |      |  X  |    X  |               |
 --------------------------------------------------------------
 |                 |     |      |     |       |               |

@@ Line 40: / Line 40: @@
 Q0000003.wav
 Q0000003.pv
 etc.
@@ Line 240: / Line 239: @@
 . Both 48 ground-truth midis (or their variants in pv or wav) and 2797 wavs (or their variants in pv or mid) are pull together for such task.
 . Only top-ten candidates are returned. The performance evaluation is based on precision of the returned set.

← Older revision		Revision as of 15:48, 6 September 2006
Line 231:		Line 231:
	The two step calling format proposed above by Stephen Downie is going to be used. So people can use different type of directories for indexing and query steps. For example: one can index test database in midi format and use the querry database in wav format or in anyone of the other 3 formats. Please indicate in your README file which format you want to use for each step.		The two step calling format proposed above by Stephen Downie is going to be used. So people can use different type of directories for indexing and query steps. For example: one can index test database in midi format and use the querry database in wav format or in anyone of the other 3 formats. Please indicate in your README file which format you want to use for each step.
	The answer file format above will be used.		The answer file format above will be used.
		+
		+	==Comment by Rainer Typke (20060906)==
		+	Just an observation: The "match beginning" property is yet another difference to the Symbolic Melodic Similarity task 2 (Mix/Karaoke), where you cannot assume that the match will be at the beginning. (I'm gonna submit the same algorithm anyways, just with slightly altered parameters).

← Older revision		Revision as of 02:26, 6 September 2006
Line 226:		Line 226:
	2. This is up to Mert Bay who should be in charge of running everyone's program. Mery, what do you think?<br>		2. This is up to Mert Bay who should be in charge of running everyone's program. Mery, what do you think?<br>
	3. As mentioned earlier, the "match beginning" CAN be assumed in your systems since the corpus is thus designed (and its too late to change it. IMHO, for real-world systems, as long as we can cut each midi into phrases in advance, every queries are "match from beginning".		3. As mentioned earlier, the "match beginning" CAN be assumed in your systems since the corpus is thus designed (and its too late to change it. IMHO, for real-world systems, as long as we can cut each midi into phrases in advance, every queries are "match from beginning".
		+
		+	==Comment by Mert Bay (20060905)==
		+	2. No need to read subdirectories. Every file will be renamed with a unique id. For both tasks, there will a test directory to index and a query directory. Each test and query directories will have 3 different versions. One directory will contain all files in midi format, one will contain wav format and one for pv format. The file ids of the same song in different directories will be the same but with different extensions.
		+	The two step calling format proposed above by Stephen Downie is going to be used. So people can use different type of directories for indexing and query steps. For example: one can index test database in midi format and use the querry database in wav format or in anyone of the other 3 formats. Please indicate in your README file which format you want to use for each step.
		+	The answer file format above will be used.

@@ Line 192: / Line 192: @@
 ==Comment by Xiao Wu==
 I find that all singers were singing from the beginning of corresponding songs. This coincidence CANNOT be applied as prior knowledge of the submitted system, right?
 ==Comment by Christian Sailer (2)==
@@ Line 200: / Line 203: @@
 . Is really everybody ok with reading directories? For me, it's perfectly ok.<br>
 . For Task 2, should the ground truth midis be synthesized? I think yes, as this avoids the introduction of a midi classifier interface in programs that are written and submitted for classifying wavs.<br>
 ==Comment by Xiao Wu==
@@ Line 213: / Line 221: @@
 Thanks in advance.