Operating the FDK encoder in A2DP with variable bitrate mode
configuration resulted in unexpected encoder return value AACENC_ENCODE_ERROR.
Due to peak bitrate restriction the encoder quite often runs into requantization.
In case the bitrate constraint is not fulfilled, the so-called crash recovery is
used as final emergency step. The crash recovery reduces the overall bit
consumption considering the given number of bits to be saved. The bit difference
is extracted from audio element structures. In VBR mode the element wise bit
consumption state was not updated since there is typically no bitrate limitation
required.
The patch solves the choppy audio problems and increases audio quality for
AAC VBR encoding.
The changes in FDKaacEnc_QCMain() ensure that audio element bit info is always
updated. This is achieved by always calling FDKaacEnc_BitResRedistribution()
and FDKaacEnc_prepareBitDistribution() with maxBitsPerFrame variable as total
bits parameter.
Furthermore, VBR assumes a certain target bitrate which is used for internal
configuration and limitations. In case the peak bitrate parameter is less than
the VBR mode target bitrate the maximum of both bitrate configurations was used.
The function FDKaacEnc_AdjustVBRBitrateMode() is added to adjust the encoder
internal VBR mode to a desired target bitrate less than given peak bitrate.
It is possible that the peak bitrate is very close to the desired target bitrate.
The virtual available bitreservoir is quite low and the encoder would run quite
often into requantization with needless audio quality reduction. In such a
configuration, it is a better choice to use the CBR targeted threshold adaption
to avoid audio quality reduction.
In FDKaacEnc_Initialize(), there was already a bitResMode selection depending
on available bitreservoir for CBR. This selection will now also be used for VBR.
In case the bitResMode is AACENC_BR_MODE_FULL and VBR mode is selected, the VBR
threshold adaption is used. Otherwise, CBR threshold adaption strategy is used
and therefore, no unnecessary fill bits are written.
Bug: 161400526
Test: see bug
Change-Id: I1865f817180150da6add2623a64f1a102622784a
Added SPDX-license-identifier-Apache-2.0 to:
fuzzer/Android.bp
Added SPDX-license-identifier-Apache-2.0 legacy_by_exception_only to:
Android.bp
Bug: 68860345
Bug: 151177513
Bug: 151953481
Test: m all
Exempt-From-Owner-Approval: janitorial work
Change-Id: I518bcad1cb91b80b2bcce1726f53aadf9afc749e
Prevent error return during parsing of incomplete uniDrcGain() element. This fixes unnecessary concealment of frames for AAC in case a uniDrcGain() is provided, but the uniDrcConfig() element is not available.
Bug: 176246647
Test: atest android.media.cts.DecoderTestAacFormat android.media.cts.DecoderTestXheAac android.media.cts.DecoderTestAacDrc
Change-Id: I5ae343eaea481774c53240dbce4da0f4ff06cd1d
We have observed quality problems regarding config switching for USAC streams. Crossfading did not consider the USAC time domain limiter latency correctly. The limiter memory still contained the last part of the frame before the config change.
With this patch we were able to improve the quality by moving the limiter processing to the end of the processing chain (crossfade -> DRC -> limiter). By that we don't have to consider the limiter latency at the crossfader anymore and can resolve the quality issue.
Bug: 176246647
Test: atest android.media.cts.DecoderTestAacFormat android.media.cts.DecoderTestXheAac android.media.cts.DecoderTestAacDrc
Change-Id: I0dfd3b76ff2b0daf495ad406283f56a39982ad8f
Change-Id: I26f5da65ef8344602007e180e837820c6a25f173
The enums don't cover all possible values read from the bitstream.
This fixes undefined behaviour sanitizer errors.
Fixes: 27647/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_LIBFDK_AAC_fuzzer-5654559200116736
Fixes: 28193/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_LIBFDK_AAC_fuzzer-4901213455515648
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg