Implement trim optimization in Transformer

PiperOrigin-RevId: 584622392

libraries/test_data/src/test/assets/transformerdumps/mp4/sample_with_increasing_timestamps_320w_240h.mp4/trimOptimizedClippedAtKeyFrame.dump

@@ -0,0 +1,560 @@
+format video:
+  id = 1
+  sampleMimeType = video/avc
+  codecs = avc1.42C015
+  maxInputSize = 14839
+  width = 320
+  height = 240
+  frameRate = 59.997425
+  colorInfo:
+    colorSpace = 2
+    colorRange = 1
+    colorTransfer = 3
+    lumaBitdepth = 8
+    chromaBitdepth = 8
+  metadata = entries=[TSSE: description=null: values=[Lavf58.76.100], Mp4Timestamp: creation time=0, modification time=0, timescale=1000]
+  initializationData:
+    data = length 31, hash 4B108214
+    data = length 9, hash FBA158BB
+container metadata = entries=[TSSE: description=null: values=[Lavf58.76.100], Mp4Timestamp: creation time=0, modification time=0, timescale=1000]
+sample:
+  trackType = video
+  dataHashCode = 983000500
+  size = 13539
+  isKeyFrame = true
+  presentationTimeUs = 0
+sample:
+  trackType = video
+  dataHashCode = -1834230781
+  size = 32
+  isKeyFrame = false
+  presentationTimeUs = 16666
+sample:
+  trackType = video
+  dataHashCode = 521720738
+  size = 1534
+  isKeyFrame = false
+  presentationTimeUs = 33333
+sample:
+  trackType = video
+  dataHashCode = 722836039
+  size = 123
+  isKeyFrame = false
+  presentationTimeUs = 50000
+sample:
+  trackType = video
+  dataHashCode = -1702585381
+  size = 2061
+  isKeyFrame = false
+  presentationTimeUs = 66666
+sample:
+  trackType = video
+  dataHashCode = -365856396
+  size = 147
+  isKeyFrame = false
+  presentationTimeUs = 83333
+sample:
+  trackType = video
+  dataHashCode = 1258185334
+  size = 2534
+  isKeyFrame = false
+  presentationTimeUs = 100000
+sample:
+  trackType = video
+  dataHashCode = -179623006
+  size = 87
+  isKeyFrame = false
+  presentationTimeUs = 116666
+sample:
+  trackType = video
+  dataHashCode = -541393824
+  size = 2762
+  isKeyFrame = false
+  presentationTimeUs = 133333
+sample:
+  trackType = video
+  dataHashCode = -1912932514
+  size = 57
+  isKeyFrame = false
+  presentationTimeUs = 150000
+sample:
+  trackType = video
+  dataHashCode = 485634444
+  size = 2833
+  isKeyFrame = false
+  presentationTimeUs = 166666
+sample:
+  trackType = video
+  dataHashCode = 570625802
+  size = 189
+  isKeyFrame = false
+  presentationTimeUs = 183333
+sample:
+  trackType = video
+  dataHashCode = 1819668957
+  size = 3153
+  isKeyFrame = false
+  presentationTimeUs = 200000
+sample:
+  trackType = video
+  dataHashCode = 1004398066
+  size = 104
+  isKeyFrame = false
+  presentationTimeUs = 216666
+sample:
+  trackType = video
+  dataHashCode = 2087741113
+  size = 2304
+  isKeyFrame = false
+  presentationTimeUs = 233333
+sample:
+  trackType = video
+  dataHashCode = -419782502
+  size = 222
+  isKeyFrame = false
+  presentationTimeUs = 250000
+sample:
+  trackType = video
+  dataHashCode = -1867110345
+  size = 2306
+  isKeyFrame = false
+  presentationTimeUs = 266666
+sample:
+  trackType = video
+  dataHashCode = 1908323737
+  size = 257
+  isKeyFrame = false
+  presentationTimeUs = 283333
+sample:
+  trackType = video
+  dataHashCode = 884063337
+  size = 2201
+  isKeyFrame = false
+  presentationTimeUs = 300000
+sample:
+  trackType = video
+  dataHashCode = -1308458590
+  size = 174
+  isKeyFrame = false
+  presentationTimeUs = 316666
+sample:
+  trackType = video
+  dataHashCode = -1686938678
+  size = 2524
+  isKeyFrame = false
+  presentationTimeUs = 333333
+sample:
+  trackType = video
+  dataHashCode = -1372845971
+  size = 171
+  isKeyFrame = false
+  presentationTimeUs = 350000
+sample:
+  trackType = video
+  dataHashCode = 1130876644
+  size = 2306
+  isKeyFrame = false
+  presentationTimeUs = 366666
+sample:
+  trackType = video
+  dataHashCode = 1707671352
+  size = 188
+  isKeyFrame = false
+  presentationTimeUs = 383333
+sample:
+  trackType = video
+  dataHashCode = 300233313
+  size = 2529
+  isKeyFrame = false
+  presentationTimeUs = 400000
+sample:
+  trackType = video
+  dataHashCode = -1284013406
+  size = 182
+  isKeyFrame = false
+  presentationTimeUs = 416666
+sample:
+  trackType = video
+  dataHashCode = -2088617828
+  size = 2047
+  isKeyFrame = false
+  presentationTimeUs = 433333
+sample:
+  trackType = video
+  dataHashCode = 2116374999
+  size = 259
+  isKeyFrame = false
+  presentationTimeUs = 450000
+sample:
+  trackType = video
+  dataHashCode = -2123019940
+  size = 2234
+  isKeyFrame = false
+  presentationTimeUs = 466666
+sample:
+  trackType = video
+  dataHashCode = 1901454757
+  size = 138
+  isKeyFrame = false
+  presentationTimeUs = 483333
+sample:
+  trackType = video
+  dataHashCode = 1576638059
+  size = 2088
+  isKeyFrame = false
+  presentationTimeUs = 500000
+sample:
+  trackType = video
+  dataHashCode = 1120133924
+  size = 151
+  isKeyFrame = false
+  presentationTimeUs = 516666
+sample:
+  trackType = video
+  dataHashCode = 264118578
+  size = 2235
+  isKeyFrame = false
+  presentationTimeUs = 533333
+sample:
+  trackType = video
+  dataHashCode = 64254117
+  size = 164
+  isKeyFrame = false
+  presentationTimeUs = 550000
+sample:
+  trackType = video
+  dataHashCode = -1000078879
+  size = 2231
+  isKeyFrame = false
+  presentationTimeUs = 566666
+sample:
+  trackType = video
+  dataHashCode = 286919946
+  size = 123
+  isKeyFrame = false
+  presentationTimeUs = 583333
+sample:
+  trackType = video
+  dataHashCode = -320312658
+  size = 2303
+  isKeyFrame = false
+  presentationTimeUs = 600000
+sample:
+  trackType = video
+  dataHashCode = 1057750590
+  size = 175
+  isKeyFrame = false
+  presentationTimeUs = 616666
+sample:
+  trackType = video
+  dataHashCode = 1961415074
+  size = 2165
+  isKeyFrame = false
+  presentationTimeUs = 633333
+sample:
+  trackType = video
+  dataHashCode = 667267023
+  size = 260
+  isKeyFrame = false
+  presentationTimeUs = 650000
+sample:
+  trackType = video
+  dataHashCode = 979033489
+  size = 1924
+  isKeyFrame = false
+  presentationTimeUs = 666666
+sample:
+  trackType = video
+  dataHashCode = -1974473017
+  size = 286
+  isKeyFrame = false
+  presentationTimeUs = 683333
+sample:
+  trackType = video
+  dataHashCode = -962519103
+  size = 1992
+  isKeyFrame = false
+  presentationTimeUs = 700000
+sample:
+  trackType = video
+  dataHashCode = -1312094075
+  size = 204
+  isKeyFrame = false
+  presentationTimeUs = 716666
+sample:
+  trackType = video
+  dataHashCode = 2068151127
+  size = 1826
+  isKeyFrame = false
+  presentationTimeUs = 733333
+sample:
+  trackType = video
+  dataHashCode = -1531967506
+  size = 284
+  isKeyFrame = false
+  presentationTimeUs = 750000
+sample:
+  trackType = video
+  dataHashCode = -778066699
+  size = 1940
+  isKeyFrame = false
+  presentationTimeUs = 766666
+sample:
+  trackType = video
+  dataHashCode = -1219952117
+  size = 129
+  isKeyFrame = false
+  presentationTimeUs = 783333
+sample:
+  trackType = video
+  dataHashCode = -1218204223
+  size = 1947
+  isKeyFrame = false
+  presentationTimeUs = 800000
+sample:
+  trackType = video
+  dataHashCode = -1816247511
+  size = 147
+  isKeyFrame = false
+  presentationTimeUs = 816666
+sample:
+  trackType = video
+  dataHashCode = 299686318
+  size = 2066
+  isKeyFrame = false
+  presentationTimeUs = 833333
+sample:
+  trackType = video
+  dataHashCode = -1520242765
+  size = 185
+  isKeyFrame = false
+  presentationTimeUs = 850000
+sample:
+  trackType = video
+  dataHashCode = -1702498409
+  size = 2159
+  isKeyFrame = false
+  presentationTimeUs = 866666
+sample:
+  trackType = video
+  dataHashCode = 345202950
+  size = 189
+  isKeyFrame = false
+  presentationTimeUs = 883333
+sample:
+  trackType = video
+  dataHashCode = 220746796
+  size = 2098
+  isKeyFrame = false
+  presentationTimeUs = 900000
+sample:
+  trackType = video
+  dataHashCode = -32341189
+  size = 159
+  isKeyFrame = false
+  presentationTimeUs = 916666
+sample:
+  trackType = video
+  dataHashCode = -1838476361
+  size = 1914
+  isKeyFrame = false
+  presentationTimeUs = 933333
+sample:
+  trackType = video
+  dataHashCode = -1322093590
+  size = 99
+  isKeyFrame = false
+  presentationTimeUs = 950000
+sample:
+  trackType = video
+  dataHashCode = -1391064751
+  size = 2168
+  isKeyFrame = false
+  presentationTimeUs = 966666
+sample:
+  trackType = video
+  dataHashCode = 1479204931
+  size = 129
+  isKeyFrame = false
+  presentationTimeUs = 983333
+sample:
+  trackType = video
+  dataHashCode = 1131230500
+  size = 2327
+  isKeyFrame = false
+  presentationTimeUs = 1000000
+sample:
+  trackType = video
+  dataHashCode = -393815961
+  size = 160
+  isKeyFrame = false
+  presentationTimeUs = 1016666
+sample:
+  trackType = video
+  dataHashCode = -242739025
+  size = 2136
+  isKeyFrame = false
+  presentationTimeUs = 1033333
+sample:
+  trackType = video
+  dataHashCode = 65238903
+  size = 163
+  isKeyFrame = false
+  presentationTimeUs = 1050000
+sample:
+  trackType = video
+  dataHashCode = 1720840922
+  size = 2043
+  isKeyFrame = false
+  presentationTimeUs = 1066666
+sample:
+  trackType = video
+  dataHashCode = -1006231050
+  size = 178
+  isKeyFrame = false
+  presentationTimeUs = 1083333
+sample:
+  trackType = video
+  dataHashCode = 1742965952
+  size = 2022
+  isKeyFrame = false
+  presentationTimeUs = 1100000
+sample:
+  trackType = video
+  dataHashCode = -971065365
+  size = 240
+  isKeyFrame = false
+  presentationTimeUs = 1116666
+sample:
+  trackType = video
+  dataHashCode = 1757434551
+  size = 1887
+  isKeyFrame = false
+  presentationTimeUs = 1133333
+sample:
+  trackType = video
+  dataHashCode = 1501849116
+  size = 252
+  isKeyFrame = false
+  presentationTimeUs = 1150000
+sample:
+  trackType = video
+  dataHashCode = 825501977
+  size = 1816
+  isKeyFrame = false
+  presentationTimeUs = 1166666
+sample:
+  trackType = video
+  dataHashCode = -1616223509
+  size = 246
+  isKeyFrame = false
+  presentationTimeUs = 1183333
+sample:
+  trackType = video
+  dataHashCode = 457119646
+  size = 1817
+  isKeyFrame = false
+  presentationTimeUs = 1200000
+sample:
+  trackType = video
+  dataHashCode = -1382929639
+  size = 146
+  isKeyFrame = false
+  presentationTimeUs = 1216666
+sample:
+  trackType = video
+  dataHashCode = -1580853131
+  size = 1929
+  isKeyFrame = false
+  presentationTimeUs = 1233333
+sample:
+  trackType = video
+  dataHashCode = 1758706551
+  size = 196
+  isKeyFrame = false
+  presentationTimeUs = 1250000
+sample:
+  trackType = video
+  dataHashCode = 207289556
+  size = 2154
+  isKeyFrame = false
+  presentationTimeUs = 1266666
+sample:
+  trackType = video
+  dataHashCode = -981284942
+  size = 182
+  isKeyFrame = false
+  presentationTimeUs = 1283333
+sample:
+  trackType = video
+  dataHashCode = 855103964
+  size = 2144
+  isKeyFrame = false
+  presentationTimeUs = 1300000
+sample:
+  trackType = video
+  dataHashCode = 380479426
+  size = 90
+  isKeyFrame = false
+  presentationTimeUs = 1316666
+sample:
+  trackType = video
+  dataHashCode = -1677996152
+  size = 2005
+  isKeyFrame = false
+  presentationTimeUs = 1333333
+sample:
+  trackType = video
+  dataHashCode = 1516852008
+  size = 156
+  isKeyFrame = false
+  presentationTimeUs = 1350000
+sample:
+  trackType = video
+  dataHashCode = -1602805193
+  size = 1772
+  isKeyFrame = false
+  presentationTimeUs = 1366666
+sample:
+  trackType = video
+  dataHashCode = -1720426556
+  size = 162
+  isKeyFrame = false
+  presentationTimeUs = 1383333
+sample:
+  trackType = video
+  dataHashCode = -1392260423
+  size = 1865
+  isKeyFrame = false
+  presentationTimeUs = 1400000
+sample:
+  trackType = video
+  dataHashCode = -1842432151
+  size = 151
+  isKeyFrame = false
+  presentationTimeUs = 1416666
+sample:
+  trackType = video
+  dataHashCode = -537063215
+  size = 1848
+  isKeyFrame = false
+  presentationTimeUs = 1433333
+sample:
+  trackType = video
+  dataHashCode = 2089388394
+  size = 206
+  isKeyFrame = false
+  presentationTimeUs = 1450000
+sample:
+  trackType = video
+  dataHashCode = -1761777019
+  size = 1934
+  isKeyFrame = false
+  presentationTimeUs = 1466666
+sample:
+  trackType = video
+  dataHashCode = 235471194
+  size = 119
+  isKeyFrame = false
+  presentationTimeUs = 1483333
+released = true

libraries/test_data/src/test/assets/transformerdumps/wav/sample.wav/clipped.dump

@@ -0,0 +1,31 @@
+format audio:
+  averageBitrate = 131072
+  sampleMimeType = audio/mp4a-latm
+  channelCount = 1
+  sampleRate = 44100
+  pcmEncoding = 2
+sample:
+  trackType = audio
+  dataHashCode = -8136122
+  size = 8820
+  isKeyFrame = true
+  presentationTimeUs = 0
+sample:
+  trackType = audio
+  dataHashCode = 1750866613
+  size = 8820
+  isKeyFrame = true
+  presentationTimeUs = 100000
+sample:
+  trackType = audio
+  dataHashCode = -1100753636
+  size = 8820
+  isKeyFrame = true
+  presentationTimeUs = 200000
+sample:
+  trackType = audio
+  dataHashCode = 507833230
+  size = 8820
+  isKeyFrame = true
+  presentationTimeUs = 300000
+released = true

libraries/transformer/src/main/java/androidx/media3/transformer/ExportResult.java

@@ -16,7 +16,9 @@
 package androidx.media3.transformer;
 
 import static androidx.media3.common.util.Assertions.checkArgument;
+import static java.lang.annotation.ElementType.TYPE_USE;
 
+import androidx.annotation.IntDef;
 import androidx.annotation.Nullable;
 import androidx.media3.common.C;
 import androidx.media3.common.ColorInfo;
@@ -24,6 +26,10 @@ import androidx.media3.common.MediaItem;
 import androidx.media3.common.util.UnstableApi;
 import com.google.common.collect.ImmutableList;
 import com.google.errorprone.annotations.CanIgnoreReturnValue;
+import java.lang.annotation.Documented;
+import java.lang.annotation.Retention;
+import java.lang.annotation.RetentionPolicy;
+import java.lang.annotation.Target;
 import java.util.List;
 import java.util.Objects;
 import org.checkerframework.checker.nullness.qual.MonotonicNonNull;
@@ -46,6 +52,7 @@ public final class ExportResult {
     private int width;
     private int videoFrameCount;
     @Nullable private String videoEncoderName;
+    private @OptimizationResult int optimizationResult;
     @Nullable private ExportException exportException;
 
     /** Creates a builder. */
@@ -192,6 +199,21 @@ public final class ExportResult {
       return this;
     }
 
+    /**
+     * Sets {@link OptimizationResult} to indicate an optimization as been successful, or has failed
+     * and normal export proceeded instead.
+     *
+     * <p>The default value is {@link #OPTIMIZATION_NONE}.
+     *
+     * @param optimizationResult The {@link OptimizationResult}.
+     * @return This {@link Builder}.
+     */
+    @CanIgnoreReturnValue
+    public Builder setOptimizationResult(@OptimizationResult int optimizationResult) {
+      this.optimizationResult = optimizationResult;
+      return this;
+    }
+
     /** Sets the {@link ExportException} that caused the export to fail. */
     @CanIgnoreReturnValue
     public Builder setExportException(@Nullable ExportException exportException) {
@@ -215,6 +237,7 @@ public final class ExportResult {
           width,
           videoFrameCount,
           videoEncoderName,
+          optimizationResult,
           exportException);
     }
 
@@ -263,6 +286,38 @@ public final class ExportResult {
     }
   }
 
+  /**
+   * Specifies the result of an optimized operation, such as {@link
+   * Transformer.Builder#experimentalSetTrimOptimizationEnabled}. One of {@link #OPTIMIZATION_NONE},
+   * {@link #OPTIMIZATION_SUCCEEDED}, {@link #OPTIMIZATION_FAILED_NO_VIDEO_TRACK_TO_TRIM} or {@link
+   * #OPTIMIZATION_FAILED_EXTRACTION_FAILED}.
+   */
+  @Documented
+  @Retention(RetentionPolicy.SOURCE)
+  @Target(TYPE_USE)
+  @IntDef({
+    OPTIMIZATION_NONE,
+    OPTIMIZATION_SUCCEEDED,
+    OPTIMIZATION_FAILED_NO_VIDEO_TRACK_TO_TRIM,
+    OPTIMIZATION_FAILED_EXTRACTION_FAILED
+  })
+  @interface OptimizationResult {}
+
+  /** No optimizations were applied since none were requested. */
+  public static final int OPTIMIZATION_NONE = 0;
+
+  /** The optimization was successfully applied. */
+  public static final int OPTIMIZATION_SUCCEEDED = 1;
+
+  /** The trim optimization failed because there was no video track. Normal export proceeded. */
+  public static final int OPTIMIZATION_FAILED_NO_VIDEO_TRACK_TO_TRIM = 2;
+
+  /**
+   * The optimization failed because mp4 metadata extraction failed (possibly because the file
+   * wasn't an mp4 file). Normal export proceeded.
+   */
+  public static final int OPTIMIZATION_FAILED_EXTRACTION_FAILED = 3;
+
   /** The list of {@linkplain ProcessedInput processed inputs}. */
   public final ImmutableList<ProcessedInput> processedInputs;
 
@@ -306,6 +361,9 @@ public final class ExportResult {
   /** The name of the video encoder used, or {@code null} if none were used. */
   @Nullable public final String videoEncoderName;
 
+  /** The result of any requested optimizations. */
+  public final @OptimizationResult int optimizationResult;
+
   /**
    * The {@link ExportException} that caused the export to fail, or {@code null} if the export was a
    * success.
@@ -326,6 +384,7 @@ public final class ExportResult {
       int width,
       int videoFrameCount,
       @Nullable String videoEncoderName,
+      @OptimizationResult int optimizationResult,
       @Nullable ExportException exportException) {
     this.processedInputs = processedInputs;
     this.durationMs = durationMs;
@@ -340,6 +399,7 @@ public final class ExportResult {
     this.width = width;
     this.videoFrameCount = videoFrameCount;
     this.videoEncoderName = videoEncoderName;
+    this.optimizationResult = optimizationResult;
     this.exportException = exportException;
   }
 

libraries/transformer/src/main/java/androidx/media3/transformer/Transformer.java

@@ -20,6 +20,9 @@ import static androidx.media3.common.util.Assertions.checkArgument;
 import static androidx.media3.common.util.Assertions.checkNotNull;
 import static androidx.media3.common.util.Assertions.checkState;
 import static androidx.media3.transformer.Composition.HDR_MODE_EXPERIMENTAL_FORCE_INTERPRET_HDR_AS_SDR;
+import static androidx.media3.transformer.ExportResult.OPTIMIZATION_FAILED_EXTRACTION_FAILED;
+import static androidx.media3.transformer.ExportResult.OPTIMIZATION_FAILED_NO_VIDEO_TRACK_TO_TRIM;
+import static androidx.media3.transformer.TransmuxTranscodeHelper.buildNewCompositionWithClipTimes;
 import static java.lang.annotation.ElementType.TYPE_USE;
 
 import android.content.Context;
@@ -99,6 +102,7 @@ public final class Transformer {
     private boolean removeAudio;
     private boolean removeVideo;
     private boolean flattenForSlowMotion;
+    private boolean trimOptimizationEnabled;
     private ListenerSet<Transformer.Listener> listeners;
     private AssetLoader.@MonotonicNonNull Factory assetLoaderFactory;
     private AudioMixer.Factory audioMixerFactory;
@@ -138,6 +142,7 @@ public final class Transformer {
       this.videoEffects = transformer.videoEffects;
       this.removeAudio = transformer.removeAudio;
       this.removeVideo = transformer.removeVideo;
+      this.trimOptimizationEnabled = transformer.trimOptimizationEnabled;
       this.listeners = transformer.listeners;
       this.assetLoaderFactory = transformer.assetLoaderFactory;
       this.audioMixerFactory = transformer.audioMixerFactory;
@@ -282,6 +287,39 @@ public final class Transformer {
       return this;
     }
 
+    // TODO: b/304476154 - Support audio and progress updates in trim optimization.
+    /**
+     * Sets whether to attempt to optimize trims from the start of the {@link EditedMediaItem} by
+     * transcoding as little of the file as possible and transmuxing the rest.
+     *
+     * <p>This optimization has the following limitations:
+     *
+     * <ul>
+     *   <li>Only supported for single-asset (i.e. only one {@link EditedMediaItem} in the whole
+     *       {@link Composition}) exports of mp4 files.
+     *   <li>Not guaranteed to work with any effects.
+     *   <li>Video track only (removes audio from the file).
+     *   <li>Progress updates will be unavailable.
+     * </ul>
+     *
+     * <p>{@link ExportResult#optimizationResult} will indicate whether the optimization was
+     * applied.
+     *
+     * <p>This process relies on the given {@linkplain #setEncoderFactory EncoderFactory} providing
+     * the right encoder level and profiles when transcoding, so that the transcoded and transmuxed
+     * segments of the file can be stitched together. If the file segments can't be stitched
+     * together, the {@linkplain #start(Composition, String) export operation} will throw an
+     * exception.
+     *
+     * @param enabled Whether to enable trim optimization.
+     * @return This builder.
+     */
+    @CanIgnoreReturnValue
+    public Builder experimentalSetTrimOptimizationEnabled(boolean enabled) {
+      trimOptimizationEnabled = enabled;
+      return this;
+    }
+
     /**
      * @deprecated Use {@link #addListener(Listener)}, {@link #removeListener(Listener)} or {@link
      *     #removeAllListeners()} instead.
@@ -497,6 +535,7 @@ public final class Transformer {
           removeAudio,
           removeVideo,
           flattenForSlowMotion,
+          trimOptimizationEnabled,
           listeners,
           assetLoaderFactory,
           audioMixerFactory,
@@ -668,7 +707,9 @@ public final class Transformer {
     TRANSFORMER_STATE_REMUX_PROCESSED_VIDEO,
     TRANSFORMER_STATE_PROCESS_REMAINING_VIDEO,
     TRANSFORMER_STATE_PROCESS_AUDIO,
-    TRANSFORMER_STATE_COPY_OUTPUT
+    TRANSFORMER_STATE_COPY_OUTPUT,
+    TRANSFORMER_STATE_TRANSCODE_VIDEO_START,
+    TRANSFORMER_STATE_TRANSMUX_REMAINING_VIDEO
   })
   private @interface TransformerState {}
 
@@ -677,6 +718,8 @@ public final class Transformer {
   private static final int TRANSFORMER_STATE_PROCESS_REMAINING_VIDEO = 2;
   private static final int TRANSFORMER_STATE_PROCESS_AUDIO = 3;
   private static final int TRANSFORMER_STATE_COPY_OUTPUT = 4;
+  private static final int TRANSFORMER_STATE_TRANSCODE_VIDEO_START = 5;
+  private static final int TRANSFORMER_STATE_TRANSMUX_REMAINING_VIDEO = 6;
 
   private final Context context;
   private final TransformationRequest transformationRequest;
@@ -685,6 +728,7 @@ public final class Transformer {
   private final boolean removeAudio;
   private final boolean removeVideo;
   private final boolean flattenForSlowMotion;
+  private final boolean trimOptimizationEnabled;
   private final ListenerSet<Transformer.Listener> listeners;
   @Nullable private final AssetLoader.Factory assetLoaderFactory;
   private final AudioMixer.Factory audioMixerFactory;
@@ -704,10 +748,12 @@ public final class Transformer {
   private @MonotonicNonNull String outputFilePath;
   private @MonotonicNonNull String oldFilePath;
   private @TransformerState int transformerState;
-  private ExportResumeHelper.@MonotonicNonNull ResumeMetadata resumeMetadata;
+  private TransmuxTranscodeHelper.@MonotonicNonNull ResumeMetadata resumeMetadata;
-  private @MonotonicNonNull ListenableFuture<ExportResumeHelper.ResumeMetadata>
+  private @MonotonicNonNull ListenableFuture<TransmuxTranscodeHelper.ResumeMetadata>
       getResumeMetadataFuture;
   private @MonotonicNonNull ListenableFuture<Void> copyOutputFuture;
+  private @MonotonicNonNull ListenableFuture<Mp4MetadataInfo> getMp4MetadataInfoFuture;
+  private @MonotonicNonNull Mp4MetadataInfo mp4MetadataInfo;
 
   private Transformer(
       Context context,
@@ -717,6 +763,7 @@ public final class Transformer {
       boolean removeAudio,
       boolean removeVideo,
       boolean flattenForSlowMotion,
+      boolean trimOptimizationEnabled,
       ListenerSet<Listener> listeners,
       @Nullable AssetLoader.Factory assetLoaderFactory,
       AudioMixer.Factory audioMixerFactory,
@@ -734,6 +781,7 @@ public final class Transformer {
     this.removeAudio = removeAudio;
     this.removeVideo = removeVideo;
     this.flattenForSlowMotion = flattenForSlowMotion;
+    this.trimOptimizationEnabled = trimOptimizationEnabled;
     this.listeners = listeners;
     this.assetLoaderFactory = assetLoaderFactory;
     this.audioMixerFactory = audioMixerFactory;
@@ -868,12 +916,17 @@ public final class Transformer {
    * @throws IllegalStateException If an export is already in progress.
    */
   public void start(Composition composition, String path) {
+    verifyApplicationThread();
     initialize(composition, path);
-    startInternal(
+    if (!trimOptimizationEnabled || isMultiAsset()) {
-        composition,
+      startInternal(
-        new MuxerWrapper(path, muxerFactory, componentListener, MuxerWrapper.MUXER_MODE_DEFAULT),
+          composition,
-        componentListener,
+          new MuxerWrapper(path, muxerFactory, componentListener, MuxerWrapper.MUXER_MODE_DEFAULT),
-        /* initialTimestampOffsetUs= */ 0);
+          componentListener,
+          /* initialTimestampOffsetUs= */ 0);
+    } else {
+      transcodeVideoBeforeFirstSyncSampleAfterTrimStartTime();
+    }
   }
 
   /**
@@ -1026,6 +1079,10 @@ public final class Transformer {
    *   <li>The output is an MP4 file.
    * </ul>
    *
+   * <p>Note that export optimizations (such as {@linkplain
+   * Builder#experimentalSetTrimOptimizationEnabled trim optimization}) will not be applied upon
+   * resumption.
+   *
    * @param composition The {@link Composition} to resume export.
    * @param outputFilePath The path to the output file. This must be different from the output path
    *     of the cancelled export.
@@ -1060,13 +1117,13 @@ public final class Transformer {
   private void remuxProcessedVideo() {
     transformerState = TRANSFORMER_STATE_REMUX_PROCESSED_VIDEO;
     getResumeMetadataFuture =
-        ExportResumeHelper.getResumeMetadataAsync(
+        TransmuxTranscodeHelper.getResumeMetadataAsync(
             context, checkNotNull(oldFilePath), checkNotNull(composition));
     Futures.addCallback(
         getResumeMetadataFuture,
-        new FutureCallback<ExportResumeHelper.ResumeMetadata>() {
+        new FutureCallback<TransmuxTranscodeHelper.ResumeMetadata>() {
           @Override
-          public void onSuccess(ExportResumeHelper.ResumeMetadata resumeMetadata) {
+          public void onSuccess(TransmuxTranscodeHelper.ResumeMetadata resumeMetadata) {
             // If there is no video track to remux or the last sync sample is actually the first
             // sample, then start the normal Export.
             if (resumeMetadata.lastSyncSampleTimestampUs == C.TIME_UNSET
@@ -1085,7 +1142,7 @@ public final class Transformer {
                     MuxerWrapper.MUXER_MODE_MUX_PARTIAL_VIDEO);
 
             startInternal(
-                ExportResumeHelper.createVideoOnlyComposition(
+                TransmuxTranscodeHelper.createVideoOnlyComposition(
                     oldFilePath,
                     /* clippingEndPositionUs= */ resumeMetadata.lastSyncSampleTimestampUs),
                 checkNotNull(remuxingMuxerWrapper),
@@ -1105,7 +1162,7 @@ public final class Transformer {
   private void processRemainingVideo() {
     transformerState = TRANSFORMER_STATE_PROCESS_REMAINING_VIDEO;
     Composition videoOnlyComposition =
-        ExportResumeHelper.buildUponComposition(
+        TransmuxTranscodeHelper.buildUponComposition(
             checkNotNull(composition),
             /* removeAudio= */ true,
             /* removeVideo= */ false,
@@ -1125,7 +1182,7 @@ public final class Transformer {
     transformerState = TRANSFORMER_STATE_PROCESS_AUDIO;
 
     startInternal(
-        ExportResumeHelper.createAudioTranscodeAndVideoTransmuxComposition(
+        TransmuxTranscodeHelper.createAudioTranscodeAndVideoTransmuxComposition(
             checkNotNull(composition), checkNotNull(outputFilePath)),
         new MuxerWrapper(
             checkNotNull(oldFilePath),
@@ -1140,7 +1197,7 @@ public final class Transformer {
   private void copyOutput() {
     transformerState = TRANSFORMER_STATE_COPY_OUTPUT;
     copyOutputFuture =
-        ExportResumeHelper.copyFileAsync(
+        TransmuxTranscodeHelper.copyFileAsync(
             new File(checkNotNull(oldFilePath)), new File(checkNotNull(outputFilePath)));
 
     Futures.addCallback(
@@ -1161,6 +1218,102 @@ public final class Transformer {
         applicationHandler::post);
   }
 
+  private void transcodeVideoBeforeFirstSyncSampleAfterTrimStartTime() {
+    transformerState = TRANSFORMER_STATE_TRANSCODE_VIDEO_START;
+    MediaItem firstMediaItem =
+        checkNotNull(composition).sequences.get(0).editedMediaItems.get(0).mediaItem;
+    long trimStartTimeUs = firstMediaItem.clippingConfiguration.startPositionUs;
+    getMp4MetadataInfoFuture =
+        TransmuxTranscodeHelper.getMp4MetadataInfo(
+            context,
+            checkNotNull(firstMediaItem.localConfiguration).uri.toString(),
+            trimStartTimeUs);
+    Futures.addCallback(
+        getMp4MetadataInfoFuture,
+        new FutureCallback<Mp4MetadataInfo>() {
+          @Override
+          public void onSuccess(Mp4MetadataInfo mp4MetadataInfo) {
+            if (mp4MetadataInfo.firstSyncSampleTimestampUsAfterTimeUs == C.TIME_UNSET) {
+              exportResultBuilder.setOptimizationResult(OPTIMIZATION_FAILED_NO_VIDEO_TRACK_TO_TRIM);
+              processFullInput();
+              return;
+            }
+            if (mp4MetadataInfo.firstSyncSampleTimestampUsAfterTimeUs == trimStartTimeUs) {
+              Transformer.this.composition =
+                  buildNewCompositionWithClipTimes(
+                      composition,
+                      trimStartTimeUs,
+                      firstMediaItem.clippingConfiguration.endPositionUs,
+                      mp4MetadataInfo.durationUs,
+                      /* startsAtKeyFrame= */ true);
+              processFullInput();
+              return;
+            }
+
+            Transformer.this.mp4MetadataInfo = mp4MetadataInfo;
+            Composition trancodeComposition =
+                buildNewCompositionWithClipTimes(
+                    composition,
+                    trimStartTimeUs,
+                    mp4MetadataInfo.firstSyncSampleTimestampUsAfterTimeUs,
+                    mp4MetadataInfo.durationUs,
+                    /* startsAtKeyFrame= */ false);
+
+            // TODO: b/304476154 - Check for cases where we shouldTranscode anyway and proceed with
+            //  normal export instead.
+            remuxingMuxerWrapper =
+                new MuxerWrapper(
+                    checkNotNull(outputFilePath),
+                    muxerFactory,
+                    componentListener,
+                    MuxerWrapper.MUXER_MODE_MUX_PARTIAL_VIDEO);
+            startInternal(
+                trancodeComposition,
+                remuxingMuxerWrapper,
+                componentListener,
+                /* initialTimestampOffsetUs= */ 0);
+          }
+
+          @Override
+          public void onFailure(Throwable t) {
+            exportResultBuilder.setOptimizationResult(OPTIMIZATION_FAILED_EXTRACTION_FAILED);
+            processFullInput();
+          }
+        },
+        applicationHandler::post);
+  }
+
+  private void transmuxRemainingVideo() {
+    transformerState = TRANSFORMER_STATE_TRANSMUX_REMAINING_VIDEO;
+    // TODO: b/304476154 - check original file format against transcode file format here to fail
+    //  fast if necessary.
+    MediaItem firstMediaItem =
+        checkNotNull(composition).sequences.get(0).editedMediaItems.get(0).mediaItem;
+    long trimStartTimeUs = firstMediaItem.clippingConfiguration.startPositionUs;
+    long trimEndTimeUs = firstMediaItem.clippingConfiguration.endPositionUs;
+    checkNotNull(mp4MetadataInfo);
+    Composition transmuxComposition =
+        buildNewCompositionWithClipTimes(
+            composition,
+            mp4MetadataInfo.firstSyncSampleTimestampUsAfterTimeUs,
+            trimEndTimeUs,
+            mp4MetadataInfo.durationUs,
+            /* startsAtKeyFrame= */ true);
+    checkNotNull(remuxingMuxerWrapper);
+    remuxingMuxerWrapper.changeToAppendVideoMode();
+    startInternal(
+        transmuxComposition,
+        remuxingMuxerWrapper,
+        componentListener,
+        /* initialTimestampOffsetUs= */ mp4MetadataInfo.firstSyncSampleTimestampUsAfterTimeUs
+            - trimStartTimeUs);
+  }
+
+  private boolean isMultiAsset() {
+    return checkNotNull(composition).sequences.size() > 1
+        || composition.sequences.get(0).editedMediaItems.size() > 1;
+  }
+
   private void verifyApplicationThread() {
     if (Looper.myLooper() != looper) {
       throw new IllegalStateException("Transformer is accessed on the wrong thread.");
@@ -1173,7 +1326,6 @@ public final class Transformer {
       ComponentListener componentListener,
       long initialTimestampOffsetUs) {
     checkArgument(composition.effects.audioProcessors.isEmpty());
-    verifyApplicationThread();
     checkState(transformerInternal == null, "There is already an export in progress.");
     TransformationRequest transformationRequest = this.transformationRequest;
     if (composition.hdrMode != Composition.HDR_MODE_KEEP_HDR) {
@@ -1209,7 +1361,7 @@ public final class Transformer {
             debugViewProvider,
             clock,
             initialTimestampOffsetUs,
-            /* matchInitializationData= */ false);
+            /* matchInitializationData= */ trimOptimizationEnabled);
     transformerInternal.start();
   }
 
@@ -1258,6 +1410,11 @@ public final class Transformer {
         processAudio();
       } else if (transformerState == TRANSFORMER_STATE_PROCESS_AUDIO) {
         copyOutput();
+      } else if (transformerState == TRANSFORMER_STATE_TRANSCODE_VIDEO_START) {
+        transmuxRemainingVideo();
+      } else if (transformerState == TRANSFORMER_STATE_TRANSMUX_REMAINING_VIDEO) {
+        exportResultBuilder.setOptimizationResult(ExportResult.OPTIMIZATION_SUCCEEDED);
+        onExportCompletedWithSuccess();
       } else {
         onExportCompletedWithSuccess();
       }

libraries/transformer/src/main/java/androidx/media3/transformer/TransmuxTranscodeHelper.java

@@ -36,8 +36,8 @@ import java.io.OutputStream;
 import java.util.ArrayList;
 import java.util.List;
 
-/** Utility methods for resuming an export. */
+/** Utility methods used in transmux-transcode exports. */
-/* package */ final class ExportResumeHelper {
+/* package */ final class TransmuxTranscodeHelper {
 
   /** Provides metadata required to resume an export. */
   public static final class ResumeMetadata {
@@ -59,7 +59,59 @@ import java.util.List;
     }
   }
 
-  private ExportResumeHelper() {}
+  public static ListenableFuture<Mp4MetadataInfo> getMp4MetadataInfo(
+      Context context, String filePath, long timeUs) {
+    SettableFuture<Mp4MetadataInfo> mp4MetadataInfoSettableFuture = SettableFuture.create();
+    new Thread("TransmuxTranscodeHelper:Mp4MetadataInfo") {
+      @Override
+      public void run() {
+        try {
+          mp4MetadataInfoSettableFuture.set(Mp4MetadataInfo.create(context, filePath, timeUs));
+        } catch (Exception ex) {
+          mp4MetadataInfoSettableFuture.setException(ex);
+        }
+      }
+    }.start();
+    return mp4MetadataInfoSettableFuture;
+  }
+
+  public static Composition buildNewCompositionWithClipTimes(
+      Composition oldComposition,
+      long startTimeUs,
+      long endTimeUs,
+      long mediaDurationUs,
+      boolean startsAtKeyFrame) {
+    EditedMediaItem firstEditedMediaItem = oldComposition.sequences.get(0).editedMediaItems.get(0);
+
+    MediaItem.ClippingConfiguration clippingConfiguration =
+        new MediaItem.ClippingConfiguration.Builder()
+            .setStartPositionUs(startTimeUs)
+            .setEndPositionUs(endTimeUs)
+            .setStartsAtKeyFrame(startsAtKeyFrame)
+            .build();
+
+    MediaItem mediaItem =
+        firstEditedMediaItem
+            .mediaItem
+            .buildUpon()
+            .setClippingConfiguration(clippingConfiguration)
+            .build();
+    EditedMediaItem editedMediaItem =
+        firstEditedMediaItem
+            .buildUpon()
+            .setMediaItem(mediaItem)
+            .setDurationUs(mediaDurationUs)
+            // TODO: b/304476154 - Support audio in trim optimization.
+            .setRemoveAudio(true)
+            .build();
+
+    return oldComposition
+        .buildUpon()
+        .setSequences(ImmutableList.of(new EditedMediaItemSequence(editedMediaItem)))
+        .build();
+  }
+
+  private TransmuxTranscodeHelper() {}
 
   /**
    * Returns a video only {@link Composition} from the given {@code filePath} and {@code
@@ -94,7 +146,7 @@ import java.util.List;
   public static Composition createAudioTranscodeAndVideoTransmuxComposition(
       Composition composition, String videoFilePath) {
     Composition audioOnlyComposition =
-        ExportResumeHelper.buildUponComposition(
+        TransmuxTranscodeHelper.buildUponComposition(
             checkNotNull(composition),
             /* removeAudio= */ false,
             /* removeVideo= */ true,
@@ -203,7 +255,7 @@ import java.util.List;
   public static ListenableFuture<ResumeMetadata> getResumeMetadataAsync(
       Context context, String filePath, Composition composition) {
     SettableFuture<ResumeMetadata> resumeMetadataSettableFuture = SettableFuture.create();
-    new Thread("ExportResumeHelper:ResumeMetadata") {
+    new Thread("TransmuxTranscodeHelper:ResumeMetadata") {
       @Override
       public void run() {
         try {
@@ -256,7 +308,7 @@ import java.util.List;
   /** Copies {@link File} content from source to destination asynchronously. */
   public static ListenableFuture<Void> copyFileAsync(File source, File destination) {
     SettableFuture<Void> copyFileSettableFuture = SettableFuture.create();
-    new Thread("ExportResumeHelper:CopyFile") {
+    new Thread("TransmuxTranscodeHelper:CopyFile") {
       @Override
       public void run() {
         if (copyFileSettableFuture.isCancelled()) {

libraries/transformer/src/test/java/androidx/media3/transformer/MediaItemExportTest.java

@@ -152,6 +152,87 @@ public final class MediaItemExportTest {
             /* modifications...= */ "clipped"));
   }
 
+  @Test
+  public void start_trimOptimizationEnabled_clippingConfigurationUnset_outputMatchesOriginal()
+      throws Exception {
+    Transformer transformer =
+        createTransformerBuilder(muxerFactory, /* enableFallback= */ false)
+            .experimentalSetTrimOptimizationEnabled(true)
+            .build();
+    MediaItem mediaItem =
+        new MediaItem.Builder()
+            .setUri(ASSET_URI_PREFIX + FILE_AUDIO_VIDEO_INCREASING_TIMESTAMPS_15S)
+            .build();
+
+    transformer.start(mediaItem, outputDir.newFile().getPath());
+    ExportResult exportResult = TransformerTestRunner.runLooper(transformer);
+
+    assertThat(exportResult.optimizationResult).isEqualTo(ExportResult.OPTIMIZATION_NONE);
+    // Asserts against file generated when experimentalSetTrimOptimizationEnabled is set to false.
+    DumpFileAsserts.assertOutput(
+        context,
+        muxerFactory.getCreatedMuxer(),
+        getDumpFileName(/* originalFileName= */ FILE_AUDIO_VIDEO_INCREASING_TIMESTAMPS_15S));
+  }
+
+  @Test
+  public void start_trimOptimizationEnabled_withClippingStartAtKeyFrame_completesSuccessfully()
+      throws Exception {
+    Transformer transformer =
+        createTransformerBuilder(muxerFactory, /* enableFallback= */ false)
+            .experimentalSetTrimOptimizationEnabled(true)
+            .build();
+    MediaItem mediaItem =
+        new MediaItem.Builder()
+            .setUri(ASSET_URI_PREFIX + FILE_AUDIO_VIDEO_INCREASING_TIMESTAMPS_15S)
+            .setClippingConfiguration(
+                new MediaItem.ClippingConfiguration.Builder()
+                    .setStartPositionMs(12_500)
+                    .setEndPositionMs(14_000)
+                    .build())
+            .build();
+
+    transformer.start(mediaItem, outputDir.newFile().getPath());
+    ExportResult exportResult = TransformerTestRunner.runLooper(transformer);
+
+    assertThat(exportResult.optimizationResult).isEqualTo(ExportResult.OPTIMIZATION_NONE);
+    // TODO: b/304476154 - When trim optimization supports audio, remove trim optimization specific
+    //   file and use the pre-existing clipped file made from normal export path.
+    DumpFileAsserts.assertOutput(
+        context,
+        muxerFactory.getCreatedMuxer(),
+        getDumpFileName(
+            /* originalFileName= */ FILE_AUDIO_VIDEO_INCREASING_TIMESTAMPS_15S,
+            /* modifications...= */ "trimOptimizedClippedAtKeyFrame"));
+  }
+
+  @Test
+  public void start_trimOptimizationEnabled_fileNotMp4_fallbackToNormalExport() throws Exception {
+    Transformer transformer =
+        createTransformerBuilder(muxerFactory, /* enableFallback= */ false)
+            .experimentalSetTrimOptimizationEnabled(true)
+            .build();
+    MediaItem mediaItem =
+        new MediaItem.Builder()
+            .setUri(ASSET_URI_PREFIX + FILE_AUDIO_RAW)
+            .setClippingConfiguration(
+                new MediaItem.ClippingConfiguration.Builder()
+                    .setStartPositionUs(500_000)
+                    .setEndPositionUs(900_000)
+                    .build())
+            .build();
+
+    transformer.start(mediaItem, outputDir.newFile().getPath());
+    ExportResult exportResult = TransformerTestRunner.runLooper(transformer);
+
+    assertThat(exportResult.optimizationResult).isNotEqualTo(ExportResult.OPTIMIZATION_SUCCEEDED);
+    assertThat(exportResult.optimizationResult).isNotEqualTo(ExportResult.OPTIMIZATION_NONE);
+    DumpFileAsserts.assertOutput(
+        context,
+        muxerFactory.getCreatedMuxer(),
+        getDumpFileName(/* originalFileName= */ FILE_AUDIO_RAW, /* modifications...= */ "clipped"));
+  }
+
   @Test
   public void start_withSubtitlesVideoOnly_completesSuccessfully() throws Exception {
     Transformer transformer =

libraries/transformer/src/test/java/androidx/media3/transformer/SequenceExportTest.java

@@ -29,6 +29,7 @@ import static androidx.media3.transformer.TestUtil.createPitchChangingAudioProce
 import static androidx.media3.transformer.TestUtil.createTransformerBuilder;
 import static androidx.media3.transformer.TestUtil.getDumpFileName;
 import static androidx.media3.transformer.TestUtil.removeEncodersAndDecoders;
+import static com.google.common.truth.Truth.assertThat;
 
 import android.content.Context;
 import androidx.media3.common.MediaItem;
@@ -177,6 +178,56 @@ public final class SequenceExportTest {
             "transmux"));
   }
 
+  @Test
+  public void
+      start_trimOptimizationEnabled_concatenateClippedMediaItemsWithTransmux_completesSuccessfully()
+          throws Exception {
+    Transformer transformer =
+        createTransformerBuilder(muxerFactory, /* enableFallback= */ false)
+            .experimentalSetTrimOptimizationEnabled(true)
+            .build();
+    MediaItem.ClippingConfiguration clippingConfiguration1 =
+        new MediaItem.ClippingConfiguration.Builder()
+            .setStartPositionMs(0) // Corresponds to key frame.
+            .setEndPositionMs(500)
+            .build();
+    MediaItem mediaItem1 =
+        new MediaItem.Builder()
+            .setUri(ASSET_URI_PREFIX + FILE_AUDIO_VIDEO_INCREASING_TIMESTAMPS_15S)
+            .setClippingConfiguration(clippingConfiguration1)
+            .build();
+    EditedMediaItem editedMediaItem1 = new EditedMediaItem.Builder(mediaItem1).build();
+    MediaItem.ClippingConfiguration clippingConfiguration2 =
+        new MediaItem.ClippingConfiguration.Builder()
+            .setStartPositionMs(12_500) // Corresponds to key frame.
+            .setEndPositionMs(14_000)
+            .build();
+    MediaItem mediaItem2 =
+        new MediaItem.Builder()
+            .setUri(ASSET_URI_PREFIX + FILE_AUDIO_VIDEO_INCREASING_TIMESTAMPS_15S)
+            .setClippingConfiguration(clippingConfiguration2)
+            .build();
+    EditedMediaItem editedMediaItem2 = new EditedMediaItem.Builder(mediaItem2).build();
+    Composition composition =
+        new Composition.Builder(new EditedMediaItemSequence(editedMediaItem1, editedMediaItem2))
+            .setTransmuxAudio(true)
+            .setTransmuxVideo(true)
+            .build();
+
+    transformer.start(composition, outputDir.newFile().getPath());
+    ExportResult exportResult = TransformerTestRunner.runLooper(transformer);
+
+    assertThat(exportResult.optimizationResult).isEqualTo(ExportResult.OPTIMIZATION_NONE);
+    DumpFileAsserts.assertOutput(
+        context,
+        muxerFactory.getCreatedMuxer(),
+        getDumpFileName(
+            /* originalFileName= */ FILE_AUDIO_VIDEO_INCREASING_TIMESTAMPS_15S,
+            /* modifications...= */ "clipped",
+            "clipped",
+            "transmux"));
+  }
+
   @Test
   public void concatenateAudioAndSilence_withTransmuxVideo_completesSuccessfully()
       throws Exception {