Changes

Page history
Update 2025 07 30 IVAS BASOP PC report authored by fotopoulou's avatar fotopoulou
Show whitespace changes
Inline Side-by-side
Meetings/Reports/2025-07-30-IVAS-BASOP-PC-report.md
View page @ 12f5445b
......@@ -2,20 +2,20 @@
* Presence
* Cadence: @ghanekar (), @ahomkar (), @patils ()
* Dolby: @bruhnst (), @sbrow (), @jtorr (), @tyagiri ()
* Ericsson: @janssontoftg (), @svedberg (), @kinuthia (), @norvell (), @demirkanik ()
* FhG: @multrus (), @fotopoulou<sup>1</sup> (x), @kiene (), @sagnowski (), @doehla (), @tamarapu (), @weckbecker (), @mullerfa (), @jander (), @naghibzadeht ()
* Dolby: @bruhnst (), @sbrow (), @jtorr (), @tyagiri (x)
* Ericsson: @janssontoftg (), @svedberg (x), @kinuthia (x), @norvell (), @demirkanik ()
* FhG: @multrus (x), @fotopoulou<sup>1</sup> (x), @kiene (x), @sagnowski (), @doehla (x), @tamarapu (x), @weckbecker (), @mullerfa (), @jander (), @naghibzadeht ()
* Huawei: @suhua (), @gaoyuy (), Isaac ()
* Nokia: @laaksonenl (), @laitinenmik (), @pihlajakuja (), @ramoa (), @vasilache (), @lintervo (), @paulusj (), @pajunen (), @mates ()
* NTT: @moriya (), @ohbah (), @sekine (), @sasaki (), Noboru ()
* Orange: @ragot (), @lefort (), @emerit ()
* Panasonic: @ehara (), @haradaa (), @srikanthn ()
* Philips: @szczerba (), @devansh.kandpal (), @oomenw ()
* Nokia: @laaksonenl (x), @laitinenmik (), @pihlajakuja (), @ramoa (x), @vasilache (x), @lintervo (), @paulusj (), @pajunen (), @mates ()
* NTT: @moriya (x), @ohbah (), @sekine (), @sasaki (), Noboru ()
* Orange: @ragot (x), @lefort (x), @emerit ()
* Panasonic: @ehara (x), @haradaa (), @srikanthn ()
* Philips: @szczerba (x), @devansh.kandpal (x), @oomenw ()
* Qualcomm: @schevciwa (), @dsinder (), @genovesea ()
* VoiceAge: @jelinek (), @malenovsky (), @vaclav (), @vaillancour ()
* VoiceAge: @jelinek (), @malenovsky (x), @vaclav (), @vaillancour (x)
* Xiaomi: @wangbbin1 ()
* <sup>1</sup> DSP Solutions GmbH representing Fraunhofer IIS
* Agreement agenda
* Agreement agenda -\> agreed
# CI
......@@ -31,18 +31,46 @@
* \--\> Finalize outstanding comments and then merge. Will need to be merged in CI merge.
* Coordination of porting work
* Work on porting ongoing in parallel
* New runner from Dolby -\> enabled today
* please report if any issues
# Technical issues
- Work stalled due to multiple regressions for merging main-pc to main. Reported regressions from merge to main:
* #1838 [regression-merge] Decoder OMASA rateswitch JBM increased MAX_ASB_DIFF
* #1837 [regression-merge] Decoder MASA DTX BINAURAL out
* #1838 \[regression-merge\] Decoder OMASA rateswitch JBM increased MAX_ASB_DIFF
* @multrus how severe is these issues
* @vasilache didn't have a look at the code yet. How reliable is the abs diff measure for bitrate switching and JBM
* @multrus depends on the MAX_ABS_DIFF value
* @vasilache we looked at issues with MAX_ABS_DIFF\> 1000
* @doehla is time scaling identical? This is one first thing to check
* @multrus we need to decide which issues are blockers, to proceed with the merge
* @vasilache there are also some OMASA Encoder cases that are not reported yet, the differences where already large
* \-\> label as "blocker" to understand which issues prevent the merge
* #1837 \[regression-merge\] Decoder MASA DTX BINAURAL out
* @vasilache level of difference is slightly over 1000
* #1835 \[regression-merge\] Decoder: Loudness difference ivas-float-update vs. float-pc for MC BINAURAL_ROOM_IR + Headrotation output
* @multrus binaural renderer porting work was not correct. Fixed logic, but saturation and asserts. Work ongoing. Many regressions with head-rotation and reverb, not sure if it is related
* https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/issues/1834 \[regression-merge\] different levels for SBA to BINAURAL_ROOM_REVERB between ivas-float-update and float-pc
* @szczerba give priority to porting the fixes to float-pc
* @multrus is float-pc inline with main-pc, then the difference should be approx. the same
* @tyagiri new module being enabled that may introduce new issues - not tested by Ittiam
* @multrus accept differences and fix issues after merge? Not sure which strategy is best
* @szczerba some control logic changes that are missing, make these MRs available ASAP and decide tomorrow - preference would be to fix these issues
* @multrus if float-pc and main-pc are inline, then this should not block the merge. Would not add new patches at this point to avoid new possible regressions
* @szczerba it is inline
* @multrus close the ticket
* https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/issues/1833 \[regression-merge\] SBA Decoder rendering to BINAURAL/BINAURAL_ROOM_IR - max. abs. diff regression BASOP vs. Float
* https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/issues/1832 \[regression-merge\] SBA decoder in float-pc and main-pc applies different bandwidth extension than ivas-float-update
* @kiene title misleading, does not seem to be a BWE issue.
* @tyagiri This seems to be a bug fix in float code for DirAC - this causes more energy in higher frequencies. This was ported now to float-pc and caused these "regressions".
* https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/issues/1831 \[regression-merge\] Decoder-Output for BINAURAL_ROOM_REVERB with rend_config_hospital_patientroom.cfg very different in BASOP vs. Float
* @multrus regression MAX_ABS_DIFF look better after merge, however, many cases for BINAURAL_ROOM_REVERB. Would be good to understand if this is a separate issue
* @szczerba will have a look
* @kiene rendering path differences between main-pc and float-pc
* @multrus high likelihood that something is not in sync
* https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/issues/1828 \[regression-merge\] LTV MASA 2TC 16 kHz in 48 kHz out at 64 kbps has a spike in main-pc
* @multrus this doesn't seem to be a blocker
* decide in tomorrow's meeting whether we proceed with merge
- How to deal with down-mix exceeding of 16-bit range?
- #1793, #1520, #1773 (last three on McMASA).
- @multrus: How severe is #1793?
......@@ -52,7 +80,10 @@
- \--\> Check how severe the issues are @vasilache.
- Cleanup of \_o vs \_sat operators, where Overflow is ignored.
- FhG currently working on lib_com (@bauerfa)
- Any volunteer to help?
- Any volunteer to help? -\> doesn't seem to be any
* @tyagiri can help with the review. I need to understand better
* @svedberg is this for the EVS code too?
* @multrus yes. In a lot of places overflow bit is set but not handled
- Clean up of switches
- Set of switches to accept was agreed in https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/wikis/Meetings/Reports/2025-07-16%20IVAS%20BASOP%20report.
- Accept after merge of main-pc into main.
......@@ -91,7 +122,7 @@
* !1281 \[CI\] add ltv version of encoder dmx comparison
* Other
* !1941 MCT issue in debug mode
* !1933 possible fix for #1766, mld goes from 23.75 to 11.72 on the mono output [allow-regression]
* !1933 possible fix for #1766, mld goes from 23.75 to 11.72 on the mono output \[allow-regression\]
* !1931 Resolve "Replace carry/overflow-using 32-bit operators by 64-bit version"
* !1927 Resolve "Usage of W_mac_32_32() + ONE_IN_Q30"
* !1895 Resolve #1808 "BASOP Asserts for BINAURAL_ROOM_REVERB output on main-pc"
......
......