International organisation for standardisation
- Bu səhifədəki naviqasiya:
- 4General Interest Presentations
- 4.2Overview of Professional Archive MAF
| Score diff.
|
Best ref. Codec. |
Type | |
|
Item | |||
|
51,11 |
AMR-WB+ |
S |
te1_mg54_speech |
|
48,00 |
AMR-WB+ |
S |
Wedding_speech |
|
45,00 |
AMR-WB+ |
S |
te19 |
|
42,22 |
AMR-WB+ |
S |
carrot_speech |
|
28,56 |
HE-AAC |
M |
phi7 |
|
21,56 |
HE-AAC |
M |
trilogy |
|
21,33 |
AMR-WB+ |
SOM |
HarryPotter |
|
20,89 |
HE-AAC |
M |
salvation |
|
20,44 |
AMR-WB+ |
SOM |
te16_fe49 |
|
15,56 |
HE-AAC |
M |
dongwoo |
|
14,67 |
HE-AAC |
SOM |
SpeechOverMusic_3 |
|
10,22 |
HE-AAC |
SOM |
twinkle_ff51 |
Eunmi Oh, Samsung, presented
|
m15462 |
Test item selection for Unified Speech and Audio Coding |
Eunmi Oh Miyoung Kim |
This contribution used a process similar to the one proposed by Philips at the last meeting, shown here:
-
max (AMR-WB+ - HE-AAC v2): to improve content dependency of reference codecs -
max (HE-AAC v2 - AMR-WB+): to improve content dependency of reference codecs -
max (VC): to maintain the best performance of virtual codec -
min (VC): to improve worst case behavior for virtual codec
As additional criterion, the contribution notes that test items should represent signals addressing the following issues:
-
to consider envisioned applications such mobile streaming and broadcasting where various types of contents exist -
to achieve content robustness: equally well handle speech and music signals -
to maintain the best performance of existing codecs -
to improve coding efficiency at low bitrates
Eunmi Oh noted that bitrate and mono or stereo have a major impact on the items selected. Kristofer Kjörling, Dolby, noted that selecting based on maximum differences for the mixed category might not be the best criteria since it could be that both codecs that comprise VC have good scores, or have bad scores and in both of those cases the difference of scores is small.
The operating points tested were16 kb/s mono and 24 kb/s stereo, and the items selected were:
|
|
Category | ||
|
Criteria |
Speech |
Music |
Speech over Music |
|
max (AMR - AAC) |
es02 |
brahms |
lion |
|
max (AAC - AMR) |
Wedding_ms_speech |
dongwoo |
phi6 |
|
max (VC) |
te19 |
Music_2 |
speechOverMusic_3 |
|
min (VC) |
es01 |
Music_4 |
twinkle_ff51 |
Markus Multrus, FhG, presented
|
m15427 |
Proposal for Test Item Selection for CfP on Unified Speech and Audio Coding |
Markus Multrus Max Neuenforf Ralf Geiger Bernhard Grill |
The contribution presents listening tests for 12 kb/s, 16 kb/s/ and 24 kb/s mono operating points. The worst-scoring items for each coder comprising VC were noted, without regard to category. The number of “hits” for each item was summed up and the items with the most hits were selected.
Based on these criteria, the following list of test items per category was selected:
|
Speech |
Music |
Mixed |
|
Green_speech |
Music_5 |
HarryPotter |
|
te19 |
phi7 |
SpeechOverMusic_2 |
|
es01 |
Trilogy |
|
|
te1_mg54_speech |
sc03 |
|
Johannes Boehm, Thomson, presented
|
m15465 |
Proposal for item selection, CfP Unified Speech and Audio Coding |
Johannes Boehm |
The contribution did a number of simplifications of the selection process:
-
Consider only at VC scores -
Consider only 20 kb/s stereo operating point. The presenter noted that this is below the current operating for MPEG technology.
Given this, the process was to select
-
2 items that were Min(VC) -
2 items that were Max(VC)
Only two listeners participated in the test. The following items were selected:
|
|
Group | ||
|
Criterion |
M |
S |
SM |
|
Min(VC) |
phi2 |
Wedding_speech |
HarryPotter |
|
Min2(VC) |
Music_4 |
Arirang_speech |
SpeechOverMusic_3 |
|
Max(VC) |
brahms |
te19 |
twinkle_ff51 |
|
Max2(VC) |
Music_1 |
es03 |
te16_fe49 |
Discussion
The AhG recommends that there be a break-out group during the MPEG week to continue to discuss item selection, including:
-
Selection process -
Operating modes (bitrate and mono or stereo) at which listening data is available and the number of listeners at each operating mode -
Training procedure and training item selection
Anisse Taleb, Ericsson, proposed that there might be different items at different operating points. Heiko Purnhagen, Dolby, noted that this makes rate-distortion analysis problematic. The Chair noted that, typically, a single set of items have been used to assess performance at different rates. Pierrick Philippe, France Telecom Research, noted that perhaps 12 items, 4 for each of the three categories could be agreed on.
Werner Oomen, Philips, provided the following summary information:
-
Test
kb/s
S/M
Tested by (subjects)
1
64k
S
-
2
32
S
-
3
24
S
Samsung (3), Philips (2)
4
24
M
FhG (3), VoiceAge(3)
5
20
S
Thomson (2), ETRI(3)
6
20
M
ETRI(3)
7
16
S
-
8
16
M
Samsung (8), FhG (3), VoiceAge(3)
9
12
M
FhG (3), VoiceAge(3)
Schuyler Quackenbush, Audio Research Labs, presented
|
m15448 |
Draft Workplan on Subjective Testing of Unified Speech and Audio Coding Proposals |
Schuyler Quackenbush |
The Contribution presented a Workplan for the test of CfP responses. This draft was reviewed, some issues were clarified and remaining open issues were highlighted. The Chair encouraged experts to carefully review this draft since a final document is needed on Friday.
4General Interest Presentations
4.1Interactive Music Demonstration ETRI
This presentation described and demonstrated an consumer product in Korea that permits the user to play not only the producer mix of a music CD, but also a user-designated mix in which the user can set the level of a number of “voices” in the music, for example main vocal, chorus or specific instruments. This product is already associated with about 30 music CDs in Korea, and over 1 million interactive music CDs have been sold in ASIA (Korea, Japan, etc.).
The physical CD media is backward-compatible with conventional CD players, but has “hidden data” on CD that enables the interactive functionality.
This technology is proposed to MPEG as a MAF for physical CD and for streaming formats. The related input documents are as following.
-
m15390, Consideration of Interactive Music Service -
m15391, Requirements for Interactive Music AF
4.2Overview of Professional Archive MAF
This presentation described the current status of the professional archive MAF. It is based on the MPEG-21 File Format. Hence the MAF supports a rich set of meta-data for the content set. It is envisioned that a flexible set of media compression tools will be supported, and currently MPEG-4 ALS and MPEG-4 AVC are supported. Tools are identified via identifier fields which are defined either in this specification or by appropriate registration authorities.
Yüklə 3,36 Mb.
Dostları ilə paylaş:
