The MP4 data in JPEG motion photos can contain multiple `video/hevc` tracks, but only the first is at a playable frame rate while the others are low-fps, high-res tracks designed for specific use-cases (not direct video playback).
ExoPlayer currently selects the unplayable track by default, because it
has a higher resolution. This change introduces a flag to
`Mp4Extractor` that results in the first video track being marked as
`ROLE_FLAG_MAIN`, and all subsequent video tracks `ROLE_FLAG_ALTERNATE`
- this then results in the playable lower-res track being selected by
default.
PiperOrigin-RevId: 589832072
This image has two video tracks in the MP4 data, one is a 'real' video
which we want to play by default, and the other is a low-fps video track
which isn't intended to be directly played, it's encoded in HEVC for
compression and decoding efficiency.
This test demonstrates ExoPlayer's current behaviour default extraction
and playback, which results in selecting the high-res, low-fps track
(actually single sample in this example), instead of playing the actual
video.
PiperOrigin-RevId: 588068908
As Opus decoders skip some bytes prior to playback during a seek, the renderer for bypass playback should send samples to the decoder even if they would be decode-only. However, the renderer should not send samples with time preceding that range. This change adds that constraint.
#minor-release
PiperOrigin-RevId: 588014983
Both the extension OPUS decoder and the OMX/C2 MediaCodec
implementations for OPUS and VORBIS decode into the channel
layout defined by VORBIS. See
https://www.xiph.org/vorbis/doc/Vorbis_I_spec.html#x1-140001.2.3
While this is technically correct for a stand-alone OPUS or VORBIS
decoder, it doesn't match the channel layout expected by Android.
See https://developer.android.com/reference/android/media/AudioFormat#channelMask
The fix is to apply the channel mapping after decoding if needed.
Also add e2e tests with audio dumps for the extension renderer,
including a new 5.1 channel test file.
Issue: google/ExoPlayer#8396
#minor-release
PiperOrigin-RevId: 588004832
Before, a translucent overlay over an opaque video would result in a
translucent output. This is not consistent with physical properties of light
(if putting a translucent object in front of an opaque object, you can't see
behind the opaque object).
Using the mixing properties from DefaultVideoCompositor.
PiperOrigin-RevId: 586636275
`fromBundle` doesn't distinguish between `FIELD_BITMAP` and `FIELD_TEXT`
being present with a null value, or being absent, so we might as well
avoid including them when the value is null.
I've separated this from a later change to add
`Cue.toSerializableBundle` which will also skip setting a bitmap value
into the `Bundle` if `this.bitmap == null`. This is partly because it
results in changes to a lot of extractor test dump files, and it's
easier to review that as a separate change.
PiperOrigin-RevId: 586626141
This change fixes a bug with seeking forward in MIDI. When seeking forward,
the progressive media period attempts to seek within the sample queue, if a
key-frame exists before the seeking position. With MIDI, however, we can
only skip Note-On and Note-Off samples and all other samples must be sent
to the MIDI decoder.
When seeking outside the sample queue, the MidiExtractor already
instructs the player to start from the beginning of the MIDI input. With
this change, only the first output sample is a key-frame, thus the
progressive media period can no longer seek within the sample queue and
is forced to seek from the MIDI input start always.
Issue: androidx/media#704
#minor-release
PiperOrigin-RevId: 584321443
When transmuxing, the `EncodedSampleExporter` maintains a queue of input
buffers that get filled with encoded data by the asset loader. The number of
buffers was limited to avoid using more and more memory if producer (asset
loader) gets far ahead of the consumer (exporter).
Previously this limit was fixed at 10 buffers, but increasing the number of
buffers can make some transmux operations much faster. Allow allocating between
a min and max number of buffers, and also set a target allocation size beyond
which new buffers can't be allocated. This allows audio formats which require
many small buffers to be processed more quickly, while preventing allocating
too much memory for hypothetical very high bitrate formats.
'Remove video' edits on local videos in particular get much faster, because
audio buffers are very short and there are lots of them. With a sample 10
minute video, a 'remove video' edit took 2 seconds (36 seconds before this
change). With a sample 1 minute removing video took 0.25 seconds after this
change (2.5 seconds before).
The speed improvement is smaller for other types of edits that retain the video
track. Transmuxing a 10 minute video retaining the video track took 26 seconds
(40 seconds before).
PiperOrigin-RevId: 583390284
This removes the flakiness of the HLS playback test as well.
Previously, this flow only worked for standalone WebVTT subtitles (7b762642db)
PiperOrigin-RevId: 582359383
The test is hidden behind the `@Ignore` annotation due to some
flakiness. However, it will be removed when the subtitle parsing is
moved to a pre-sample-queue architecture.
The test media was created with:
```shell
$ cp ../../dash/standalone-ttml/sample.xml sample.ttml
$ MP4Box -add sample.ttml sample.text.mp4
$ MP4Box -frag 10000 sample.text.mp4
$ rm sample.ttml
```
PiperOrigin-RevId: 582347113
Use renderengine's PQ to SDR tone-mapping implementation instead
of naive implementation from before.
This improves luminance on highlights, as seen in the test image.
PiperOrigin-RevId: 582318045
The dump file diff as part of this change is because using AudioGraph
means the 2nd item is automatically edited to match the AudioFormat of
the 1st item {44.1KHz mono}, rather than {48KHz stereo}.
Manually verified that for the 2nd item, data output:
* Before: 66936 bytes (16734 frames) output = 348_625us of audio.
* After: 30750 bytes (15375 frames) output = 348_639us of audio.
The small final buffer is caused by SonicAudioProcessor outputting all
pending data when EOS queued, and is WAI.
PiperOrigin-RevId: 580494578
This change applies to standalone WebVTT files linked directly from the manifest.
Since DASH only supports stand-alone IMSC1 (TTML) and WebVTT text files, this change concludes the support extension of text-based subtitle files to be parse during extraction.
PiperOrigin-RevId: 577468830
As Opus decoders skip some bytes prior to playback during a seek, the renderer for bypass playback should send samples to the decoder even if they would be decode-only.
#minor-release
PiperOrigin-RevId: 574494666
If the `Subtitle` has 'active' cues at `OutputOptions.startTimeUs`, this
change ensures these are emitted in a `CuesWithTiming` with
`CuesWithTiming.startTimeUs = OutputOptions.startTimeUs`. If
`OutputOptions.outputAllCues` is also set, then another `CuesWithTiming`
is emitted at the end that covers the 'first part' of the active cues,
and ends at `OutputOptions.startTimeUs`.
As well as adding some more tests to `LegacySubtitleUtilWebvttTest`,
this change also adds more tests for `TtmlParser` handling of
`OutputOptions`, which transitively tests the behaviour of
`LegacySubtitleUtil`.
#minor-release
PiperOrigin-RevId: 573151016
This currently only applies to subtitles muxed into mp4 segments, and
not standalone text files linked directly from the manifest.
Issue: androidx/media#288
#minor-release
PiperOrigin-RevId: 572263764
The flag is no longer used by our components and only set and checked
in a few places to guarantee compatiblity with existing renderers and
decoders that still use it.
The flag will be removed in the future due to its design limitations.
#minor-release
PiperOrigin-RevId: 571291168
The existing `Subtitle` handling code is left intact to support the
legacy post-`SampleQueue` decoding path for now.
This also includes full support for merging overlapping `CuesWithTiming`
instances, which explains the test dump file changes, and which should
resolve the following issues (if used with the
decoder-before-`SampleQueue` subtitle logic added in
5d453fcf37):
* Issue: google/ExoPlayer#10295
* Issue: google/ExoPlayer#4794
It should also help resolve Issue: androidx/media#288, but that will also require
some changes in the DASH module to enable pre-`SampleQueue` subtitle
parsing (which should happen soon).
#minor-release
PiperOrigin-RevId: 571021417
The test is hidden behind the Ignore annotation due to some flakiness just like `webvttInMp4`. However, it will be removed when the subtitle parsing is moved to a pre-sample-queue architecture.
#minor-release
PiperOrigin-RevId: 570376275
Move remote 8K file to local and trim to 320ms.
Trim done with ffmpeg:
`ffmpeg -i {remote_file} -t 0.3 -c:v copy -c:a copy 8k24fps_300ms.mp4`
PiperOrigin-RevId: 569449962
This is equivalent to the existing `scaleLargeTimestamp` method with the
following changes/improvements:
* No longer specific to timestamps (there was nothing inherently
time-specific about the logic in `scaleLargeTimestamp`, but the name
and docs suggested it shouldn't be used for non-timestamp use-cases).
* Additional 'perfect division' checks between `value` and `divisor`.
* The caller can now provide a `RoundingMode`.
* Robust against `multiplier == 0`.
* Some extra branches before falling through to (potentially lossy)
floating-point math, including trying to simplify the fraction with
greatest common divisor to reduce the chance of overflowing `long`.
This was discussed during review of 6e91f0d4c5
This change also includes some golden test file updates - these
represent a bug fix where floating-point maths had previously resulted
in a timestamp being incorrectly rounded down to the previous
microsecond. These changes are due to the 'some more branches' mentioned
above.
PiperOrigin-RevId: 564760748
Modifying dumping to not required "released" to be called.
Track index is an arbitrary value based on the order of addTrack calls.
Samples are dumped by track (rather than as soon as they are written),
so it's preferable to use a value that provides more context.
By using the track type as a key, dump files will be more deterministic
and will have more similarities when branched.
PiperOrigin-RevId: 563700982
This change makes capturing components in CapturingRenderersFactory
to dump data only they have previously captured any. This is so we can
extend the CapturingRenderersFactory with more renderers that do not
capture data in pre-existing tests and we don't have to change the
golden files.
PiperOrigin-RevId: 561973645
also improves tests to hash bitmap now that we know the bitmaps are decoded in roboelectic tests,
as well as some minor fixes in the DefaultImageDecoder
PiperOrigin-RevId: 561946073
Implement VideoCompositor support of:
* Different input and output sizes
* CompositorSettings, to customize output size based on input texture sizes
* OverlaySettings, to place an input frame in an arbitrary position on
the output frame.
Also, refactor Overlay's matrix logic to make it more reusable between
Compositor and Overlays
PiperOrigin-RevId: 561931854
This file decompresses to 3024*4032*4 bytes ~= 48
MB, which leads to OOM errors on some devices.
Use a lower resolution test image for tests, to
still test a portrait image but without hitting
OOM errors.
PiperOrigin-RevId: 557861063
In addition to this being how Alpha should be handled, tests do fail under
API 29 without this CL. Unfortunately while these tests do fail under API 29
without this CL, we currently only run this test on API 33 emulators, so we
didn't catch this failure earlier (until compositor tests on mh failed on all
devices under API 29 and succeeded on all at or over 29).
PiperOrigin-RevId: 557781757
This test uses an existing CBR test asset, and shows that (as expected)
`isSeekable = true` even in the `unknown_length` dump file.
PiperOrigin-RevId: 557455364
This effect can scale Alpha values.
This is needed for testing Compositor, and coincidentally was requested
by a partner in the past. This also will generally be more useful in
full multi-asset, where we may want to have one input have different alpha
values than another
Also, update fragment_shader_copy_es2.glsl to not throw away alpha values.
This means ThumbnailStripShaderProgram (the only place this glsl is used) will
consider input alpha and output alpha appropriately
PiperOrigin-RevId: 555915092
To ensure that for each output bitmap from the compostor, the right input
timestamps were used.
Only applied on a subset of tests to avoid needing to upload+maintain
too many files/size in the test binary, especially when it would test
duplicate behavior
PiperOrigin-RevId: 555222530
The test file decompresses to about 48 MB, which can cause an
`OutOfMemoryError` on lower end devices.
Also, a test appears to apply a rotation to this bitmap without downscaling
first, leading to exceeding a maximum texture width/height of 4096 on some
devices.
Replace the test file with one that has a quarter of the original's
width/height so that these tests can pass reliably even on lower end
devices.
PiperOrigin-RevId: 554507185
Periods that are not the last in the manifest are always ended
and non-dynamic periods are also always ended by definition.
Also, some periods may not be able to declare their final duration
accurately and we need to account for some inaccuracy. This can be
done by testing whether no further full segment within the duration
can be expected.
This is the same change as 8655429af7, just on the second call site
of readDiscontinuity. The tests didn't cover this case yet because
they never queued more than 2 items in a playlist.
PiperOrigin-RevId: 553485244
After this change, every queued bitmap is treated as an individual input stream
(like a new MediaItems).
This change merges the FrameDropTest and FrameDropPixelTest into one (while maintaining all the test cases)
- This is accomplished by generating bitmaps with timestamps on it in FrameDropTest and compare them with goldens (one may call this a pixel test, please lmk if you want this to be renamed)
- The most part of the change comes from DefaultVideoFrameProcessorVideoFrameRenderingTest. The overall working is
- We bypass the input manager
- The TestFrameGenerator generates frames based on timestamps. In this case, we generate frames with timestamps on it
- The generated frame is sent to texture output and in turn saved to bitmaps
- We then compare the generated bitmap with the goldens
PiperOrigin-RevId: 551795770
When generating silence for AudioProcessingPipeline, audio never
queued EOS downstream.
Linked to this, when silence followed an item with audio, the silence
was added to SilentAudioGenerator before the mediaItem reconfiguration
occurred. If the silence had effects, the APP would be flushed after
silence queued EOS, resetting APP.isEnded back to false, so AudioGraph
never ended.
Regression tests reproduce failure without fix, but pass with it.
PiperOrigin-RevId: 550853714
Instead of `text/x-exoplayer-cues`, we will use `application/x-media3-cues`.
The prefix should be `application` not `text` since the encoded form is arbitrary bytes, not necessarily text. The name should not reference `exoplayer`, since the Media3 Extractors (which are not part of `exoplayer`) produce this format.
PiperOrigin-RevId: 550181852
Alters the OverlayShaderProgram implementation to support rotations, however we need to apply the transformations separately in order for them to work as expected so the matrix is removed from the interface in favour of explicit methods.
Adds a rotation test to ensure this ability doesn't regress
PiperOrigin-RevId: 549890847
It is not part of the `SubtitleParser` interface's promise to give events for all changes to do with `Cue` objects. So while in the past, we would have `endTimeUs` of one Cue event being the `startTimeUs` of the next one, now we have that encoded in `durationUs` and can skip event with empty `Cues`.
PiperOrigin-RevId: 549629157
This is no longer flaky because WebVTT subtitles are decoded on the
loading thread (enabled in this test with
`defaultExractorsFactory.setTextTrackTranscodingEnabled(true)` - and
supported for WebVTT since f0f24aa0d4).
PiperOrigin-RevId: 549594291
The end-to-end test output for the overlapping SRT and SSA subtitles
is currently incorrect. They will be fixed in a future change that
updates `TextRenderer` to support this overlap.
The 'extra' samples visible in the extractor test output files are
'empty cue list' samples produced by `SsaParser`. They will go away
when this implementation is updated to remove this behaviour and rely
on `CuesWithTiming.durationUs` instead (the 'empty list' behaviour is
not required by the `SubtitleParser` interface).
PiperOrigin-RevId: 549264593
*** Original commit ***
Rollback of b69b33206e
*** Original commit ***
Mark output sample as decode-only based on start time
We currently do the same check on the input timestamps and
expect the output timestamps to match. Some codecs produce
samples with modified timestamps and the logic is a lot safer
when the comparison with the start time is done on the output
side of the codec.
Issue: google/ExoPlayer#11000
***
***
PiperOrigin-RevId: 549019403
This was done because it was deemed correct to only start at timestamp
zero when the code was originally written. However, in case of
prerolling from a keyframe, many samples will get the same timestamp,
which is not correct and interferes with downstream logic
that deals with timestamps.
PiperOrigin-RevId: 548986160
This is a step towards adding general support for overlapping
subtitles in these formats (and others), both muxed and sideloaded:
* Issue: google/ExoPlayer#10295
* Issue: google/ExoPlayer#4794
This change adds these files to the end-to-end playback tests too, but
the subtitle track is currently disabled because renderer-side subtitle
parsing causes flaky tests (due to an uncontrolled thread in
`SimpleSubtitleDecoder`). The subtitle track will be re-enabled in
a follow-up change when loading-side subtitle parsing is added (so the
tests will no longer be flaky). At this point the overlapping subtitles
**still** won't be supported end-to-end, but a second change will
resolve this will changes in `TextRenderer` - which will change the
end-to-end playback dumps to reflect the overlapping subtitles.
PiperOrigin-RevId: 548705032
Before this CL, the text with a scale span would wrap text, with because the scale wasn't taken into account when measuring the width of the overlay.
PiperOrigin-RevId: 548123626
Add Ogg ID Header and Comment Header Pages to the Ogg encapsulated Opus for offload playback. This further matches the RFC 7845 spec and provides initialization data to decoders.
PiperOrigin-RevId: 548080222
These tests allow to verify the samples sent for decoding,
when decoders are reset, which samples are dropped as decode-only
for video and which samples are sent to the AudioTrack for playback.
The test exercises all combinations of merges where audio or video
is the primary track and where audio, video or both are clipped.
PiperOrigin-RevId: 548061254
These tests allow to verify the samples sent for decoding,
when decoders are reset, which samples are dropped as decode-only
for video and which samples are sent to the AudioTrack for playback.
The test exercises all combinations of clipping transitions for
sources that are either clipped at the start, the end, or both.
PiperOrigin-RevId: 547730824
We currently only log the input buffers to the codec, but the test
will be more complete and provide more insights into the playback
behavior if we also dump more information about these input buffers
(timestamps and flags) and also all output buffers with their
timestamps and whether they are rendererd (only relevant for video).
And also use the CapturingAudioSink to output the audio data we produce.
PiperOrigin-RevId: 546898937
Some adjustments to the test to make it more correct and prevent it
from becoming flaky:
- Use separate output dump files per test setup. Once we add more data
to these files, they are not guaranteed to be the same anymore.
- Use a seek position that is actually behind the midroll as described
in the test setup.
- Change ad insertion position to ensure the ad group duration doesn't
exceed the underlying media duration.
- Add a wait for isLoading to ensure the late insertion of an ad group
happens consistently at the same processing stage.
PiperOrigin-RevId: 546825183
This avoids issues with potential number clashes with ID_UNSET.
Also fixes some further parsing where adaptation set ids are referenced
from manifest properties.
On a MediaItem change, the input Format (and Effects to apply) may be
different. Therefore the AudioProcessingPipeline must be reconfigured
to determine what processing is active, and what the AudioFormat of the
data output is. In the event that it is different, additional
AudioProcessor instances must be used to ensure the encoder will still
be able to accept the audio buffers.
PiperOrigin-RevId: 544338451
Goal of tests (SequenceExportTest) that use this media is for the
silence and the media to match exactly with audio format, however
`sample_with_increasing_timestamps.mp4` had a different sample rate.
testvid_1022ms.mp4: channel count = 2, sample rate = 44100.
PiperOrigin-RevId: 543458948
MP4 edit lists sometimes ask to start playback between two samples.
If this happens, we currently change the timestamp of the first
sample to zero to trim it (e.g. to display the first frame for a
slightly shorter period of time). However, we can't do this to audio
samples are they have an inherent duration and trimming them this
way is not possible.
#minor-release
PiperOrigin-RevId: 543420218
With the upcoming "handle format changes" CL, stereo -> mono audio
would add an AudioProcessor. Robolectric decodes output encoded data,
which crashes some AudioProcessors because the number of frames may not
be an integer.
PiperOrigin-RevId: 542568875
Audio only tests are now using RAW audio where possible, which is
passed through the Robolectric decoders/encoders, and can be handled by
the AudioProcessor instances accurately.
PiperOrigin-RevId: 541648853
Issue: When running the Transformer related test cases, the tests are flaky
because the order in which audio and video samples are interleaved seems to
differ in few instances.
Root cause: When running a transformation the sample producer (Asset loader)
and sample consumer (Sample pipeline) both runs on different thread and
theoretically there is no reason for behaviour to be deterministic because
the number of samples produced/written depends on how fast individual thread
works. So it is indeed surprising that test somehow worked deterministically in
majority of instances (may be something to do with Robolectric environment).
Solution: Since we don't expect the order of sample interleaving to be deterministic, make the dumping logic deterministic where all the video
samples will be collected and then dumped together (similarly for audio). This would mean we won't be able to see the interleaving so for that we need to
add separate test case verifying the interleaving logic only.
Pending: Test case for interleaving.
PiperOrigin-RevId: 540930871
The test transcodes four media items with distinct effects, keeping one frame
each; extracts the four frames in the produced video, and compares them against
the expected results.
PiperOrigin-RevId: 539697344
Mp4Muxer already supports writing Mp4LocationData so added that
as supported Metadata entry.
Support for more Metadata entries will be added in upcoming CLs.
PiperOrigin-RevId: 534473866
This is a pre-requisite for the Android Studio upgrade assistant to
upgrade from AGP 7.2.2 to 8.0.1, otherwise it fails and complains
this is missing.
Issue: androidx/media#409
PiperOrigin-RevId: 533463246
*** Original commit ***
Rollback of 65d5132f76
*** Original commit ***
Create InAppMuxer in transformer
To use the InAppMuxer, the client needs to pass InAppMuxer Factory.
***
***
PiperOrigin-RevId: 531470081
This change adds end-to-end Robolectric playback tests which handle
the scenario the player is playing audio via passthrough and
AudioTrack raises the ERROR_DEAD_OBJECT error upon which the player
attempts to recover by switching to another audio format.
PiperOrigin-RevId: 531180183
*** Original commit ***
Create InAppMuxer in transformer
To use the InAppMuxer, the client needs to pass InAppMuxer Factory.
***
PiperOrigin-RevId: 531056436
Previously, I assumed that SDR contents must be BT709, and HDR contents must be
BT2020. Turns out BT2020 is just wide-gamut, and SDR contents / transfers may be
represented in BT2020 color spaces.
Relax the check, so that we don't throw when valid BT2020 SMPTE 170M contents
are input into effects.
PiperOrigin-RevId: 526668347
The actual csd data contains NAL units so it should always start with
a NAL unit start code (00 00 00 01).
This issue was not caught before because in the code there was no
validation to check whether its a valid NAL unit.
PiperOrigin-RevId: 524849867
Simplify the audio encoder input timestamp calculation. The new calculation
avoids drifting by tracking the total number of bytes encoded rather than
tracking the timestamp and remainder separately, and also makes the timestamps
match the decoder output buffer timestamps.
Also switch one of the export tests that was passing through AMR samples over
to using WAVE audio. The problem with using AMR is that the compressed samples
are not necessarily an integer number of audio frames and the shadow decoder
would pass them from input to output, so the audio encoder was receiving
non-integer numbers of audio frames.
Tested by logging the timestamps at the decoder output and encoder input with
forcing transcoding audio, and verifying that after this change the audio
timestamps are no longer off by one.
PiperOrigin-RevId: 523409869
