Compare commits

..

19 Commits

Author SHA1 Message Date
Dean
dd26812a9c fix: Add pasp box to declare square pixels (1:1) for web playback
The codec string fix caused videos to appear squished on web players
like Shaka. Adding an explicit pixel aspect ratio (pasp) box with
1:1 ratio tells the player not to apply any SAR scaling.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-13 12:22:43 -08:00
Dean
b716608379 fix: Skip NAL header byte when reading SPS profile data in HlsMuxer
The SPS NAL unit format is: [NAL header, profile_idc, constraint_flags, level_idc, ...]
The code was incorrectly reading from byte 0 (NAL header, typically 0x67)
instead of byte 1 (profile_idc).

This produced invalid codec strings like `avc1.676400` instead of valid
ones like `avc1.64001f`, causing Shaka Player on web to fail with error
4032 (unable to parse codec).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-13 11:52:08 -08:00
0ecc3d8210 Merge pull request 'fix: Handle both file:// and file:/ URI prefixes' (#19) from dean/fix-file-prefix-fmp4 into main
Reviewed-on: #19
Reviewed-by: Ivan Malison <ivanmalison@gmail.com>
2026-01-06 19:41:32 +00:00
309e1e9457 Merge branch 'main' into dean/fix-file-prefix-fmp4 2026-01-06 17:38:24 +00:00
71b08e6898 Merge pull request 'Android Fmp4' (#17) from loewy/android-fmp4-normalize-timestamp-fix-fps into main
Reviewed-on: #17
2026-01-06 17:21:29 +00:00
Dean
699481f6f8 fix: Handle both file:// and file:/ URI prefixes
The previous code only stripped file:// (double slash) but some paths
come with file:/ (single slash), causing FileNotFoundException.

Fixes RAILBIRD-FRONTEND-1JH

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-06 08:45:39 -08:00
11ce9ba8f6 ensure compatability with rb chunked and fmp4, move orientation detection for rb chunked to chunked manager 2026-01-03 13:40:09 -08:00
dd9de38a7d use window manager to determine device rotation in android 2026-01-02 10:04:49 -08:00
3f5d0a2109 fix fps (alter tfhd and trun size, add logs 2026-01-02 10:04:49 -08:00
6c2319608d normalize timestamps and fix framerate metadata in init file 2026-01-02 10:04:47 -08:00
27f127fe94 Fix orientation issues 2026-01-02 10:02:51 -08:00
92b29cbd78 Write our own muxer to make hls uupload actually work 2026-01-02 10:02:51 -08:00
fb23c57a6c feat: Add fragmented MP4 (fMP4) support for Android
Implements HLS-compatible fragmented MP4 recording on Android using
AndroidX Media3 FragmentedMp4Muxer, matching the iOS implementation.

Changes:
- Add FragmentedRecordingManager for fMP4 segment output
- Add ChunkedRecorderInterface to abstract recorder implementations
- Add onInitSegmentReady callback for init segment (init.mp4)
- Update onVideoChunkReady to include segment duration
- RecordingSession now uses FragmentedRecordingManager by default

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-02 10:02:51 -08:00
8d06ab9e66 Merge pull request 'Strip file prefix ( for expo-file-system)' (#15) from loewy/stripe-file-prefix into main
Reviewed-on: #15
Reviewed-by: Ivan Malison <ivanmalison@gmail.com>
2025-12-26 17:26:42 +00:00
f6b6cfb3d5 strip file prefix 2025-12-19 12:55:41 -08:00
3ac555a2b3 Merge pull request 'Stop recording on init write failure' (#14) from loewy/stop-recording-on-init-write-failure into main
Reviewed-on: #14
2025-12-19 20:53:06 +00:00
7e1e074e0f force recording to stop on init write failure and fix silent failure 2025-12-18 13:29:31 -08:00
b269e9c493 fix silent init file write failure 2025-12-18 13:09:34 -08:00
5fe7f35127 Merge pull request 'Fix orientation Android - get orientation directly from WindowManager' (#11) from loewy/fix-android-orientation-bugs into main
Reviewed-on: #11
2025-12-17 23:20:52 +00:00
11 changed files with 318 additions and 431 deletions

View File

@@ -428,9 +428,9 @@ class CameraSession(private val context: Context, private val cameraManager: Cam
// Get actual device rotation from WindowManager since the React Native orientation hook // Get actual device rotation from WindowManager since the React Native orientation hook
// doesn't update when rotating between landscape-left and landscape-right on Android. // doesn't update when rotating between landscape-left and landscape-right on Android.
// Map device rotation to the correct orientationHint for video recording: // Map device rotation to the correct orientation for video recording.
// - Counter-clockwise (ROTATION_90) → 90° hint // Surface.ROTATION_90 = device rotated 90° CCW = phone top on LEFT = LANDSCAPE_LEFT
// - Clockwise (ROTATION_270) → 270° hint // Surface.ROTATION_270 = device rotated 90° CW = phone top on RIGHT = LANDSCAPE_RIGHT
val windowManager = context.getSystemService(Context.WINDOW_SERVICE) as WindowManager val windowManager = context.getSystemService(Context.WINDOW_SERVICE) as WindowManager
val deviceRotation = windowManager.defaultDisplay.rotation val deviceRotation = windowManager.defaultDisplay.rotation
val recordingOrientation = when (deviceRotation) { val recordingOrientation = when (deviceRotation) {
@@ -441,6 +441,8 @@ class CameraSession(private val context: Context, private val cameraManager: Cam
else -> Orientation.PORTRAIT else -> Orientation.PORTRAIT
} }
Log.i(TAG, "startRecording: orientation=${recordingOrientation.toDegrees()}° (deviceRotation=$deviceRotation)")
val recording = RecordingSession( val recording = RecordingSession(
context, context,
cameraId, cameraId,

View File

@@ -30,8 +30,15 @@ class ChunkedRecordingManager(private val encoder: MediaCodec, private val outpu
iFrameInterval: Int = 5 iFrameInterval: Int = 5
): ChunkedRecordingManager { ): ChunkedRecordingManager {
val mimeType = options.videoCodec.toMimeType() val mimeType = options.videoCodec.toMimeType()
val cameraOrientationDegrees = cameraOrientation.toDegrees() // Use cameraOrientation (from WindowManager) for rotation metadata
val recordingOrientationDegrees = (options.orientation ?: Orientation.PORTRAIT).toDegrees(); // The options.orientation from JavaScript is unreliable on Android when rotating between landscape modes
// Note: MediaMuxer.setOrientationHint() uses opposite convention from HlsMuxer's rotation matrix
// We need to invert the rotation: 90 <-> 270, while 0 and 180 stay the same
val orientationDegrees = when (cameraOrientation.toDegrees()) {
90 -> 270
270 -> 90
else -> cameraOrientation.toDegrees()
}
val (width, height) = if (cameraOrientation.isLandscape()) { val (width, height) = if (cameraOrientation.isLandscape()) {
size.height to size.width size.height to size.width
} else { } else {
@@ -55,12 +62,12 @@ class ChunkedRecordingManager(private val encoder: MediaCodec, private val outpu
format.setInteger(MediaFormat.KEY_I_FRAME_INTERVAL, iFrameInterval) format.setInteger(MediaFormat.KEY_I_FRAME_INTERVAL, iFrameInterval)
format.setInteger(MediaFormat.KEY_BIT_RATE, bitRate) format.setInteger(MediaFormat.KEY_BIT_RATE, bitRate)
Log.d(TAG, "Video Format: $format, camera orientation $cameraOrientationDegrees, recordingOrientation: $recordingOrientationDegrees") Log.d(TAG, "Video Format: $format, orientation: $orientationDegrees")
// Create a MediaCodec encoder, and configure it with our format. Get a Surface // Create a MediaCodec encoder, and configure it with our format. Get a Surface
// we can use for input and wrap it with a class that handles the EGL work. // we can use for input and wrap it with a class that handles the EGL work.
codec.configure(format, null, null, MediaCodec.CONFIGURE_FLAG_ENCODE) codec.configure(format, null, null, MediaCodec.CONFIGURE_FLAG_ENCODE)
return ChunkedRecordingManager( return ChunkedRecordingManager(
codec, outputDirectory, recordingOrientationDegrees, iFrameInterval, callbacks codec, outputDirectory, orientationDegrees, iFrameInterval, callbacks
) )
} }
} }
@@ -91,12 +98,13 @@ class ChunkedRecordingManager(private val encoder: MediaCodec, private val outpu
muxer.start() muxer.start()
} }
fun writeSample(buffer: java.nio.ByteBuffer, bufferInfo: BufferInfo) {
muxer.writeSampleData(videoTrack, buffer, bufferInfo)
}
fun finish() { fun finish() {
muxer.stop() muxer.stop()
muxer.release() muxer.release()
// Calculate duration from start time - this is approximate
// The new FragmentedRecordingManager provides accurate duration
callbacks.onVideoChunkReady(filepath, chunkIndex, null) callbacks.onVideoChunkReady(filepath, chunkIndex, null)
} }
} }
@@ -170,7 +178,7 @@ class ChunkedRecordingManager(private val encoder: MediaCodec, private val outpu
encoder.releaseOutputBuffer(index, false) encoder.releaseOutputBuffer(index, false)
return return
} }
context.muxer.writeSampleData(context.videoTrack, encodedData, bufferInfo) context.writeSample(encodedData, bufferInfo)
encoder.releaseOutputBuffer(index, false) encoder.releaseOutputBuffer(index, false)
} }
} }

View File

@@ -20,8 +20,7 @@ import java.io.File
*/ */
class FragmentedRecordingManager( class FragmentedRecordingManager(
private val encoder: MediaCodec, private val encoder: MediaCodec,
private val muxer: HlsMuxer, private val muxer: HlsMuxer
private val configuredFps: Int
) : MediaCodec.Callback(), ChunkedRecorderInterface { ) : MediaCodec.Callback(), ChunkedRecorderInterface {
companion object { companion object {
@@ -40,21 +39,18 @@ class FragmentedRecordingManager(
segmentDurationSeconds: Int = DEFAULT_SEGMENT_DURATION_SECONDS segmentDurationSeconds: Int = DEFAULT_SEGMENT_DURATION_SECONDS
): FragmentedRecordingManager { ): FragmentedRecordingManager {
val mimeType = options.videoCodec.toMimeType() val mimeType = options.videoCodec.toMimeType()
// Use cameraOrientation from Android (computed from device rotation) // Use cameraOrientation (from WindowManager) for rotation metadata
// instead of options.orientation from JS which may be stale // The options.orientation from JavaScript is unreliable on Android when rotating between landscape modes
val recordingOrientationDegrees = cameraOrientation.toDegrees() val orientationDegrees = cameraOrientation.toDegrees()
// Swap dimensions based on orientation - same logic as ChunkedRecordingManager // Swap dimensions based on camera orientation, same as ChunkedRecordingManager
// When camera is in landscape orientation, we need to swap width/height for the encoder
val (width, height) = if (cameraOrientation.isLandscape()) { val (width, height) = if (cameraOrientation.isLandscape()) {
size.height to size.width size.height to size.width
} else { } else {
size.width to size.height size.width to size.height
} }
Log.d(TAG, "Input size: ${size.width}x${size.height}, " + Log.d(TAG, "Recording: ${width}x${height}, orientation=$orientationDegrees°")
"encoder size: ${width}x${height}, " +
"orientation: $cameraOrientation ($recordingOrientationDegrees°)")
val format = MediaFormat.createVideoFormat(mimeType, width, height) val format = MediaFormat.createVideoFormat(mimeType, width, height)
val codec = MediaCodec.createEncoderByType(mimeType) val codec = MediaCodec.createEncoderByType(mimeType)
@@ -64,16 +60,18 @@ class FragmentedRecordingManager(
MediaCodecInfo.CodecCapabilities.COLOR_FormatSurface MediaCodecInfo.CodecCapabilities.COLOR_FormatSurface
) )
val effectiveFps = fps ?: 30 // Use 30fps as conservative default since many Android devices can't sustain
// higher frame rates at high resolutions. This affects:
// - Encoder: bitrate allocation and I-frame interval calculation
// - HlsMuxer: timescale for accurate sample durations
// The actual frame timing comes from camera timestamps regardless of this setting.
val effectiveFps = 30
format.setInteger(MediaFormat.KEY_FRAME_RATE, effectiveFps) format.setInteger(MediaFormat.KEY_FRAME_RATE, effectiveFps)
format.setInteger(MediaFormat.KEY_I_FRAME_INTERVAL, segmentDurationSeconds) format.setInteger(MediaFormat.KEY_I_FRAME_INTERVAL, segmentDurationSeconds)
format.setInteger(MediaFormat.KEY_BIT_RATE, bitRate) format.setInteger(MediaFormat.KEY_BIT_RATE, bitRate)
Log.d(TAG, "Video Format: $format, orientation: $recordingOrientationDegrees")
codec.configure(format, null, null, MediaCodec.CONFIGURE_FLAG_ENCODE) codec.configure(format, null, null, MediaCodec.CONFIGURE_FLAG_ENCODE)
// Create muxer with callbacks and orientation
val muxer = HlsMuxer( val muxer = HlsMuxer(
outputDirectory = outputDirectory, outputDirectory = outputDirectory,
callback = object : HlsMuxer.Callback { callback = object : HlsMuxer.Callback {
@@ -85,13 +83,12 @@ class FragmentedRecordingManager(
callbacks.onVideoChunkReady(file, index, durationUs) callbacks.onVideoChunkReady(file, index, durationUs)
} }
}, },
orientationDegrees = recordingOrientationDegrees orientationDegrees = orientationDegrees,
fps = effectiveFps
) )
muxer.setSegmentDuration(segmentDurationSeconds * 1_000_000L) muxer.setSegmentDuration(segmentDurationSeconds * 1_000_000L)
Log.d(TAG, "Created HlsMuxer with orientation: $recordingOrientationDegrees degrees, fps: $effectiveFps") return FragmentedRecordingManager(codec, muxer)
return FragmentedRecordingManager(codec, muxer, effectiveFps)
} }
} }
@@ -171,8 +168,7 @@ class FragmentedRecordingManager(
synchronized(this) { synchronized(this) {
Log.i(TAG, "Output format changed: $format") Log.i(TAG, "Output format changed: $format")
// Pass configured fps to muxer (not the encoder's output format fps which may differ) trackIndex = muxer.addTrack(format)
trackIndex = muxer.addTrack(format, configuredFps)
muxer.start() muxer.start()
muxerStarted = true muxerStarted = true
} }

View File

@@ -27,7 +27,8 @@ import java.nio.ByteBuffer
class HlsMuxer( class HlsMuxer(
private val outputDirectory: File, private val outputDirectory: File,
private val callback: Callback, private val callback: Callback,
private val orientationDegrees: Int = 0 private val orientationDegrees: Int = 0,
private val fps: Int = 30
) { ) {
companion object { companion object {
private const val TAG = "HlsMuxer" private const val TAG = "HlsMuxer"
@@ -41,8 +42,7 @@ class HlsMuxer(
// Configuration // Configuration
private var targetSegmentDurationUs: Long = DEFAULT_SEGMENT_DURATION_US private var targetSegmentDurationUs: Long = DEFAULT_SEGMENT_DURATION_US
private var timescale: Int = 30000 // Default, updated from format private var timescale: Int = 30000 // Default, updated in addTrack() to fps * 1000
private var configuredFps: Int = 30 // Configured fps from user, used for VUI timing
// State // State
private var state = State.UNINITIALIZED private var state = State.UNINITIALIZED
@@ -55,9 +55,14 @@ class HlsMuxer(
private var segmentStartTimeUs = -1L private var segmentStartTimeUs = -1L
private var lastPresentationTimeUs = 0L private var lastPresentationTimeUs = 0L
// Timestamp normalization - first timestamp becomes time 0 // Timestamp normalization - MediaCodec timestamps are device uptime, not starting from 0
private var firstPresentationTimeUs = -1L private var firstPresentationTimeUs = -1L
// Actual fps detection from frame timestamps
private var detectedFps: Int? = null
private var fpsDetectionSamples = mutableListOf<Long>()
private val FPS_DETECTION_SAMPLE_COUNT = 30
private enum class State { private enum class State {
UNINITIALIZED, UNINITIALIZED,
INITIALIZED, INITIALIZED,
@@ -79,13 +84,21 @@ class HlsMuxer(
* Normalizes a presentation timestamp to start from 0. * Normalizes a presentation timestamp to start from 0.
* The first timestamp received becomes time 0, and all subsequent * The first timestamp received becomes time 0, and all subsequent
* timestamps are relative to that. * timestamps are relative to that.
*
* This is necessary because MediaCodec timestamps are based on device uptime,
* not starting from 0. HLS players expect timestamps to start at or near 0.
*/ */
private fun normalizeTimestamp(rawPresentationTimeUs: Long): Long { private fun normalizeTimestamp(rawPresentationTimeUs: Long): Long {
if (firstPresentationTimeUs < 0) { if (firstPresentationTimeUs < 0) {
firstPresentationTimeUs = rawPresentationTimeUs firstPresentationTimeUs = rawPresentationTimeUs
Log.d(TAG, "First timestamp: ${rawPresentationTimeUs}us, normalizing to 0") Log.d(TAG, "First timestamp captured: ${rawPresentationTimeUs}us (${rawPresentationTimeUs / 1_000_000.0}s), normalizing to 0")
} }
return rawPresentationTimeUs - firstPresentationTimeUs val normalized = rawPresentationTimeUs - firstPresentationTimeUs
// Log first few normalizations to debug
if (normalized < 1_000_000) { // First second
Log.d(TAG, "Timestamp: raw=${rawPresentationTimeUs}us -> normalized=${normalized}us")
}
return normalized
} }
// ==================== Annex-B to AVCC Conversion ==================== // ==================== Annex-B to AVCC Conversion ====================
@@ -213,15 +226,18 @@ class HlsMuxer(
* Adds a track to the muxer. * Adds a track to the muxer.
* *
* @param format The MediaFormat describing the track * @param format The MediaFormat describing the track
* @param fps The configured frame rate (used for VUI timing, overrides format's fps)
* @return Track index (always 0 for now, single video track) * @return Track index (always 0 for now, single video track)
*/ */
fun addTrack(format: MediaFormat, fps: Int = 30): Int { fun addTrack(format: MediaFormat): Int {
check(state == State.UNINITIALIZED) { "addTrack() must be called before start()" } check(state == State.UNINITIALIZED) { "addTrack() must be called before start()" }
trackFormat = format trackFormat = format
configuredFps = fps
timescale = fps * 1000 // Use fps * 1000 for good precision // Use fps * 1000 as timescale for good precision (1000 timescale units per frame).
// This ensures accurate sample durations without integer truncation issues.
// Note: ffprobe may report r_frame_rate based on timescale, so the backend
// should use the explicit framesPerSecond from the API mutation, not ffprobe.
timescale = fps * 1000
state = State.INITIALIZED state = State.INITIALIZED
@@ -229,7 +245,7 @@ class HlsMuxer(
val formatHeight = try { format.getInteger(MediaFormat.KEY_HEIGHT) } catch (e: Exception) { -1 } val formatHeight = try { format.getInteger(MediaFormat.KEY_HEIGHT) } catch (e: Exception) { -1 }
Log.d(TAG, "Added track: ${format.getString(MediaFormat.KEY_MIME)}, " + Log.d(TAG, "Added track: ${format.getString(MediaFormat.KEY_MIME)}, " +
"encoder output: ${formatWidth}x${formatHeight}, " + "encoder output: ${formatWidth}x${formatHeight}, " +
"configuredFps=$configuredFps, timescale=$timescale, orientation=$orientationDegrees°") "fps=$fps, timescale=$timescale, orientation=$orientationDegrees°")
return 0 // Single track, index 0 return 0 // Single track, index 0
} }
@@ -241,16 +257,30 @@ class HlsMuxer(
check(state == State.INITIALIZED) { "Must call addTrack() before start()" } check(state == State.INITIALIZED) { "Must call addTrack() before start()" }
val format = trackFormat ?: throw IllegalStateException("No track format") val format = trackFormat ?: throw IllegalStateException("No track format")
// Create output directory if needed // Create output directory if needed, with proper error handling
if (!outputDirectory.exists()) { if (!outputDirectory.exists()) {
outputDirectory.mkdirs() val created = outputDirectory.mkdirs()
if (!created && !outputDirectory.exists()) {
throw IllegalStateException(
"Failed to create output directory: ${outputDirectory.absolutePath}. " +
"Parent exists: ${outputDirectory.parentFile?.exists()}, " +
"Parent path: ${outputDirectory.parentFile?.absolutePath}"
)
}
Log.d(TAG, "Created output directory: ${outputDirectory.absolutePath}")
} }
// Write init segment // Write init segment
val initBytes = buildInitSegment(format) val initBytes = buildInitSegment(format)
val initFile = File(outputDirectory, "init.mp4") val initFile = File(outputDirectory, "init.mp4")
FileOutputStream(initFile).use { it.write(initBytes) } FileOutputStream(initFile).use { it.write(initBytes) }
// Log frame rate metadata for debugging
val defaultSampleDuration = timescale / fps
Log.d(TAG, "Created init segment: ${initFile.absolutePath} (${initBytes.size} bytes)") Log.d(TAG, "Created init segment: ${initFile.absolutePath} (${initBytes.size} bytes)")
Log.d(TAG, "Frame rate metadata: timescale=$timescale, fps=$fps, " +
"default_sample_duration=$defaultSampleDuration (ffprobe should calculate ${timescale}/${defaultSampleDuration}=${fps}fps)")
callback.onInitSegmentReady(initFile) callback.onInitSegmentReady(initFile)
state = State.STARTED state = State.STARTED
@@ -273,13 +303,40 @@ class HlsMuxer(
} }
val isKeyFrame = (bufferInfo.flags and MediaCodec.BUFFER_FLAG_KEY_FRAME) != 0 val isKeyFrame = (bufferInfo.flags and MediaCodec.BUFFER_FLAG_KEY_FRAME) != 0
// Normalize timestamp to start from 0 (MediaCodec uses device uptime)
val presentationTimeUs = normalizeTimestamp(bufferInfo.presentationTimeUs) val presentationTimeUs = normalizeTimestamp(bufferInfo.presentationTimeUs)
// Detect actual fps from first N samples
if (detectedFps == null) {
fpsDetectionSamples.add(presentationTimeUs)
if (fpsDetectionSamples.size >= FPS_DETECTION_SAMPLE_COUNT) {
val elapsed = fpsDetectionSamples.last() - fpsDetectionSamples.first()
if (elapsed > 0) {
val actualFps = ((FPS_DETECTION_SAMPLE_COUNT - 1) * 1_000_000.0 / elapsed).toInt()
detectedFps = actualFps
if (kotlin.math.abs(actualFps - fps) > 5) {
Log.w(TAG, "Actual fps ($actualFps) differs significantly from configured fps ($fps)! " +
"This may cause processing issues if backend uses configured fps.")
} else {
Log.d(TAG, "Detected actual fps: $actualFps (configured: $fps)")
}
}
fpsDetectionSamples.clear() // Free memory
}
}
// Initialize segment start time // Initialize segment start time
if (segmentStartTimeUs < 0) { if (segmentStartTimeUs < 0) {
segmentStartTimeUs = presentationTimeUs segmentStartTimeUs = presentationTimeUs
} }
// Update duration of previous sample BEFORE finalization check
// This ensures the last sample has correct duration when segment is finalized
if (pendingSamples.isNotEmpty()) {
val lastSample = pendingSamples.last()
lastSample.durationUs = presentationTimeUs - lastSample.presentationTimeUs
}
// Check if we should finalize current segment (at keyframe boundaries) // Check if we should finalize current segment (at keyframe boundaries)
if (isKeyFrame && pendingSamples.isNotEmpty()) { if (isKeyFrame && pendingSamples.isNotEmpty()) {
val segmentDurationUs = presentationTimeUs - segmentStartTimeUs val segmentDurationUs = presentationTimeUs - segmentStartTimeUs
@@ -298,12 +355,6 @@ class HlsMuxer(
// Convert Annex-B (start codes) to AVCC (length prefixes) // Convert Annex-B (start codes) to AVCC (length prefixes)
val data = convertAnnexBToAvcc(rawData) val data = convertAnnexBToAvcc(rawData)
// Update duration of previous sample
if (pendingSamples.isNotEmpty()) {
val lastSample = pendingSamples.last()
lastSample.durationUs = presentationTimeUs - lastSample.presentationTimeUs
}
// Estimate duration (will be corrected by next sample) // Estimate duration (will be corrected by next sample)
val estimatedDurationUs = if (lastPresentationTimeUs > 0) { val estimatedDurationUs = if (lastPresentationTimeUs > 0) {
presentationTimeUs - lastPresentationTimeUs presentationTimeUs - lastPresentationTimeUs
@@ -365,6 +416,7 @@ class HlsMuxer(
val durationUs = (lastSample.presentationTimeUs - firstPts) + lastSample.durationUs val durationUs = (lastSample.presentationTimeUs - firstPts) + lastSample.durationUs
Log.d(TAG, "Created segment $segmentIndex: samples=${pendingSamples.size}, " + Log.d(TAG, "Created segment $segmentIndex: samples=${pendingSamples.size}, " +
"baseDecodeTime=${baseDecodeTimeUs}us (${baseDecodeTimeUs / 1_000_000.0}s), " +
"duration=${durationUs / 1000}ms, size=${fragmentBytes.size} bytes") "duration=${durationUs / 1000}ms, size=${fragmentBytes.size} bytes")
callback.onMediaSegmentReady(segmentFile, segmentIndex, durationUs) callback.onMediaSegmentReady(segmentFile, segmentIndex, durationUs)
@@ -378,303 +430,6 @@ class HlsMuxer(
} }
} }
// ==================== SPS VUI Timing Injection ====================
/**
* Injects VUI timing parameters into an H.264 SPS NAL unit.
* This ensures proper frame rate detection by players/decoders.
*
* The SPS from MediaCodec lacks VUI timing info, causing tools like
* ffprobe to misinterpret the frame rate.
*/
private fun injectVuiTiming(sps: ByteArray, fps: Int): ByteArray {
try {
val reader = BitReader(sps)
val writer = BitWriter()
// NAL header (1 byte: forbidden_zero_bit, nal_ref_idc, nal_unit_type)
writer.writeBits(reader.readBits(8), 8)
// profile_idc (1 byte)
val profileIdc = reader.readBits(8)
writer.writeBits(profileIdc, 8)
// constraint_set flags (1 byte)
writer.writeBits(reader.readBits(8), 8)
// level_idc (1 byte)
writer.writeBits(reader.readBits(8), 8)
// seq_parameter_set_id (ue(v))
copyExpGolomb(reader, writer)
// Profile-specific fields for High profile (100) and others
if (profileIdc == 100 || profileIdc == 110 || profileIdc == 122 ||
profileIdc == 244 || profileIdc == 44 || profileIdc == 83 ||
profileIdc == 86 || profileIdc == 118 || profileIdc == 128 ||
profileIdc == 138 || profileIdc == 139 || profileIdc == 134 ||
profileIdc == 135) {
// chroma_format_idc (ue(v))
val chromaFormatIdc = copyExpGolombAndReturn(reader, writer)
if (chromaFormatIdc == 3) {
// separate_colour_plane_flag (1 bit)
writer.writeBits(reader.readBits(1), 1)
}
// bit_depth_luma_minus8 (ue(v))
copyExpGolomb(reader, writer)
// bit_depth_chroma_minus8 (ue(v))
copyExpGolomb(reader, writer)
// qpprime_y_zero_transform_bypass_flag (1 bit)
writer.writeBits(reader.readBits(1), 1)
// seq_scaling_matrix_present_flag (1 bit)
val scalingMatrixFlag = reader.readBits(1)
writer.writeBits(scalingMatrixFlag, 1)
if (scalingMatrixFlag == 1) {
// Skip scaling lists - this is complex, just copy remaining and give up
Log.w(TAG, "SPS has scaling matrix, skipping VUI injection")
return sps
}
}
// log2_max_frame_num_minus4 (ue(v))
copyExpGolomb(reader, writer)
// pic_order_cnt_type (ue(v))
val picOrderCntType = copyExpGolombAndReturn(reader, writer)
if (picOrderCntType == 0) {
// log2_max_pic_order_cnt_lsb_minus4 (ue(v))
copyExpGolomb(reader, writer)
} else if (picOrderCntType == 1) {
// delta_pic_order_always_zero_flag (1 bit)
writer.writeBits(reader.readBits(1), 1)
// offset_for_non_ref_pic (se(v))
copySignedExpGolomb(reader, writer)
// offset_for_top_to_bottom_field (se(v))
copySignedExpGolomb(reader, writer)
// num_ref_frames_in_pic_order_cnt_cycle (ue(v))
val numRefFrames = copyExpGolombAndReturn(reader, writer)
for (i in 0 until numRefFrames) {
// offset_for_ref_frame[i] (se(v))
copySignedExpGolomb(reader, writer)
}
}
// max_num_ref_frames (ue(v))
copyExpGolomb(reader, writer)
// gaps_in_frame_num_value_allowed_flag (1 bit)
writer.writeBits(reader.readBits(1), 1)
// pic_width_in_mbs_minus1 (ue(v))
copyExpGolomb(reader, writer)
// pic_height_in_map_units_minus1 (ue(v))
copyExpGolomb(reader, writer)
// frame_mbs_only_flag (1 bit)
val frameMbsOnlyFlag = reader.readBits(1)
writer.writeBits(frameMbsOnlyFlag, 1)
if (frameMbsOnlyFlag == 0) {
// mb_adaptive_frame_field_flag (1 bit)
writer.writeBits(reader.readBits(1), 1)
}
// direct_8x8_inference_flag (1 bit)
writer.writeBits(reader.readBits(1), 1)
// frame_cropping_flag (1 bit)
val frameCroppingFlag = reader.readBits(1)
writer.writeBits(frameCroppingFlag, 1)
if (frameCroppingFlag == 1) {
// frame_crop_left_offset, right, top, bottom (ue(v) each)
copyExpGolomb(reader, writer)
copyExpGolomb(reader, writer)
copyExpGolomb(reader, writer)
copyExpGolomb(reader, writer)
}
// vui_parameters_present_flag - we'll set this to 1 and add our VUI
val originalVuiFlag = reader.readBits(1)
writer.writeBits(1, 1) // Set VUI present
// Write VUI parameters with timing info
writeVuiWithTiming(writer, fps, originalVuiFlag == 1, reader)
// Add RBSP trailing bits
writer.writeRbspTrailingBits()
val result = writer.toByteArray()
Log.d(TAG, "Injected VUI timing for ${fps}fps, SPS grew from ${sps.size} to ${result.size} bytes")
return result
} catch (e: Exception) {
Log.e(TAG, "Failed to inject VUI timing: ${e.message}, using original SPS")
return sps
}
}
/**
* Writes VUI parameters with timing info.
*/
private fun writeVuiWithTiming(writer: BitWriter, fps: Int, hadVui: Boolean, reader: BitReader) {
// aspect_ratio_info_present_flag
writer.writeBits(0, 1)
// overscan_info_present_flag
writer.writeBits(0, 1)
// video_signal_type_present_flag
writer.writeBits(0, 1)
// chroma_loc_info_present_flag
writer.writeBits(0, 1)
// timing_info_present_flag = 1
writer.writeBits(1, 1)
// num_units_in_tick (32 bits) = 1
writer.writeBits(1, 32)
// time_scale (32 bits) = fps * 2 (because each frame = 2 field counts)
writer.writeBits(fps * 2, 32)
// fixed_frame_rate_flag = 1
writer.writeBits(1, 1)
// nal_hrd_parameters_present_flag
writer.writeBits(0, 1)
// vcl_hrd_parameters_present_flag
writer.writeBits(0, 1)
// pic_struct_present_flag
writer.writeBits(0, 1)
// bitstream_restriction_flag
writer.writeBits(0, 1)
}
// ==================== Bit Manipulation Helpers ====================
/**
* Bit-level reader for parsing H.264 NAL units.
*/
private class BitReader(private val data: ByteArray) {
private var bytePos = 0
private var bitPos = 0
fun readBits(count: Int): Int {
var result = 0
for (i in 0 until count) {
if (bytePos >= data.size) throw IllegalStateException("End of data")
val bit = (data[bytePos].toInt() shr (7 - bitPos)) and 1
result = (result shl 1) or bit
bitPos++
if (bitPos == 8) {
bitPos = 0
bytePos++
}
}
return result
}
fun readExpGolomb(): Int {
var leadingZeros = 0
while (readBits(1) == 0) {
leadingZeros++
if (leadingZeros > 31) throw IllegalStateException("Invalid exp-golomb")
}
if (leadingZeros == 0) return 0
val suffix = readBits(leadingZeros)
return (1 shl leadingZeros) - 1 + suffix
}
fun readSignedExpGolomb(): Int {
val code = readExpGolomb()
return if (code % 2 == 0) -(code / 2) else (code + 1) / 2
}
}
/**
* Bit-level writer for constructing H.264 NAL units.
*/
private class BitWriter {
private val bytes = mutableListOf<Byte>()
private var currentByte = 0
private var bitPos = 0
fun writeBits(value: Int, count: Int) {
for (i in count - 1 downTo 0) {
val bit = (value shr i) and 1
currentByte = (currentByte shl 1) or bit
bitPos++
if (bitPos == 8) {
bytes.add(currentByte.toByte())
currentByte = 0
bitPos = 0
}
}
}
fun writeExpGolomb(value: Int) {
val code = value + 1
val bits = 32 - Integer.numberOfLeadingZeros(code)
// Write leading zeros
for (i in 0 until bits - 1) {
writeBits(0, 1)
}
// Write the code
writeBits(code, bits)
}
fun writeSignedExpGolomb(value: Int) {
val code = if (value <= 0) -2 * value else 2 * value - 1
writeExpGolomb(code)
}
fun writeRbspTrailingBits() {
writeBits(1, 1) // rbsp_stop_one_bit
while (bitPos != 0) {
writeBits(0, 1) // rbsp_alignment_zero_bit
}
}
fun toByteArray(): ByteArray {
// Flush remaining bits
if (bitPos > 0) {
currentByte = currentByte shl (8 - bitPos)
bytes.add(currentByte.toByte())
}
return bytes.toByteArray()
}
}
private fun copyExpGolomb(reader: BitReader, writer: BitWriter) {
val value = reader.readExpGolomb()
writer.writeExpGolomb(value)
}
private fun copyExpGolombAndReturn(reader: BitReader, writer: BitWriter): Int {
val value = reader.readExpGolomb()
writer.writeExpGolomb(value)
return value
}
private fun copySignedExpGolomb(reader: BitReader, writer: BitWriter) {
val value = reader.readSignedExpGolomb()
writer.writeSignedExpGolomb(value)
}
// ==================== Init Segment Building ==================== // ==================== Init Segment Building ====================
/** /**
@@ -684,19 +439,11 @@ class HlsMuxer(
val width = format.getInteger(MediaFormat.KEY_WIDTH) val width = format.getInteger(MediaFormat.KEY_WIDTH)
val height = format.getInteger(MediaFormat.KEY_HEIGHT) val height = format.getInteger(MediaFormat.KEY_HEIGHT)
val rawSps = format.getByteBuffer("csd-0")?.let { extractNalUnit(it) } val sps = format.getByteBuffer("csd-0")?.let { extractNalUnit(it) }
?: throw IllegalArgumentException("Missing SPS (csd-0)") ?: throw IllegalArgumentException("Missing SPS (csd-0)")
val pps = format.getByteBuffer("csd-1")?.let { extractNalUnit(it) } val pps = format.getByteBuffer("csd-1")?.let { extractNalUnit(it) }
?: throw IllegalArgumentException("Missing PPS (csd-1)") ?: throw IllegalArgumentException("Missing PPS (csd-1)")
Log.d(TAG, "Original SPS size: ${rawSps.size} bytes, PPS size: ${pps.size} bytes")
Log.d(TAG, "Original SPS hex: ${rawSps.joinToString("") { "%02x".format(it) }}")
// Inject VUI timing info into SPS using configured fps (not encoder output format fps)
val sps = injectVuiTiming(rawSps, configuredFps)
Log.d(TAG, "Modified SPS size: ${sps.size} bytes")
Log.d(TAG, "Modified SPS hex: ${sps.joinToString("") { "%02x".format(it) }}")
val output = ByteArrayOutputStream() val output = ByteArrayOutputStream()
// ftyp // ftyp
@@ -797,56 +544,91 @@ class HlsMuxer(
dos.writeShort(0) // volume (0 for video) dos.writeShort(0) // volume (0 for video)
dos.writeShort(0) // reserved dos.writeShort(0) // reserved
// Rotation matrix // Rotation matrix based on orientationDegrees
writeRotationMatrix(dos) writeRotationMatrix(dos, width, height)
// Display dimensions should be post-rotation dimensions // For 90° and 270° rotations, the display dimensions are swapped
// For 90° or 270° rotation, swap width and height // The tkhd width/height represent the final display size after rotation
val (displayWidth, displayHeight) = when (orientationDegrees) { val (displayWidth, displayHeight) = when (orientationDegrees) {
90, 270 -> height to width 90, 270 -> Pair(height, width)
else -> width to height else -> Pair(width, height)
} }
dos.writeInt(displayWidth shl 16) // width (16.16 fixed point) dos.writeInt(displayWidth shl 16) // width (16.16 fixed point)
dos.writeInt(displayHeight shl 16) // height (16.16 fixed point) dos.writeInt(displayHeight shl 16) // height (16.16 fixed point)
Log.d(TAG, "tkhd: encoded=${width}x${height}, display=${displayWidth}x${displayHeight}, rotation=$orientationDegrees") Log.d(TAG, "tkhd: encoder=${width}x${height}, display=${displayWidth}x${displayHeight}, rotation=$orientationDegrees")
return wrapBox("tkhd", output.toByteArray()) return wrapBox("tkhd", output.toByteArray())
} }
/** /**
* Writes the 3x3 transformation matrix for video rotation. * Writes the 3x3 transformation matrix for video rotation.
* The matrix is applied to rotate the video content for correct display.
*
* Matrix format in tkhd box (all values in fixed-point):
* | a b u | where a,b,c,d are 16.16 fixed-point
* | c d v | and u,v are 2.30 fixed-point (always 0)
* | x y w | x,y are 16.16, w is 2.30 (always 1.0)
*
* For rotation by θ: a=cos(θ), b=sin(θ), c=-sin(θ), d=cos(θ)
* Translation (x,y) keeps the rotated video in the visible area.
*/ */
private fun writeRotationMatrix(dos: DataOutputStream) { private fun writeRotationMatrix(dos: DataOutputStream, width: Int, height: Int) {
val one = 0x00010000 // 1.0 in 16.16 // Fixed-point constants
val negOne = 0xFFFF0000.toInt() // -1.0 in 16.16 val one = 0x00010000 // 1.0 in 16.16
val w = 0x40000000 // 1.0 in 2.30 val negOne = -0x00010000 // -1.0 in 16.16 (will be written as unsigned)
val w = 0x40000000 // 1.0 in 2.30
// For 270° device orientation (landscape-right), apply 90° CW rotation
// For 90° device orientation (landscape-left), apply 270° CW rotation
val a: Int
val b: Int
val c: Int
val d: Int
when (orientationDegrees) { when (orientationDegrees) {
90 -> { a = 0; b = negOne; c = one; d = 0 } 90 -> {
180 -> { a = negOne; b = 0; c = 0; d = negOne } // 90° rotation: x' = y, y' = -x + width
270 -> { a = 0; b = one; c = negOne; d = 0 } dos.writeInt(0) // a = 0
else -> { a = one; b = 0; c = 0; d = one } dos.writeInt(negOne) // b = -1
dos.writeInt(0) // u = 0
dos.writeInt(one) // c = 1
dos.writeInt(0) // d = 0
dos.writeInt(0) // v = 0
dos.writeInt(0) // x = 0
dos.writeInt(width shl 16) // y = width (translation)
dos.writeInt(w) // w = 1
}
180 -> {
// 180° rotation
dos.writeInt(negOne) // a = -1
dos.writeInt(0) // b = 0
dos.writeInt(0) // u = 0
dos.writeInt(0) // c = 0
dos.writeInt(negOne) // d = -1
dos.writeInt(0) // v = 0
dos.writeInt(width shl 16) // x = width (translation)
dos.writeInt(height shl 16) // y = height (translation)
dos.writeInt(w) // w = 1
}
270 -> {
// 270° rotation: x' = -y + height, y' = x
dos.writeInt(0) // a = 0
dos.writeInt(one) // b = 1
dos.writeInt(0) // u = 0
dos.writeInt(negOne) // c = -1
dos.writeInt(0) // d = 0
dos.writeInt(0) // v = 0
dos.writeInt(height shl 16) // x = height (translation)
dos.writeInt(0) // y = 0
dos.writeInt(w) // w = 1
}
else -> {
// 0° or unknown: identity matrix
dos.writeInt(one) // a = 1
dos.writeInt(0) // b = 0
dos.writeInt(0) // u = 0
dos.writeInt(0) // c = 0
dos.writeInt(one) // d = 1
dos.writeInt(0) // v = 0
dos.writeInt(0) // x = 0
dos.writeInt(0) // y = 0
dos.writeInt(w) // w = 1
}
} }
dos.writeInt(a)
dos.writeInt(b)
dos.writeInt(0) // u = 0
dos.writeInt(c)
dos.writeInt(d)
dos.writeInt(0) // v = 0
dos.writeInt(0) // tx = 0
dos.writeInt(0) // ty = 0
dos.writeInt(w) // w = 1.0
Log.d(TAG, "Rotation matrix for $orientationDegrees°")
} }
private fun buildMdiaBox(width: Int, height: Int, sps: ByteArray, pps: ByteArray): ByteArray { private fun buildMdiaBox(width: Int, height: Int, sps: ByteArray, pps: ByteArray): ByteArray {
@@ -933,7 +715,7 @@ class HlsMuxer(
private fun buildStblBox(width: Int, height: Int, sps: ByteArray, pps: ByteArray): ByteArray { private fun buildStblBox(width: Int, height: Int, sps: ByteArray, pps: ByteArray): ByteArray {
val content = ByteArrayOutputStream() val content = ByteArrayOutputStream()
content.write(buildStsdBox(width, height, sps, pps)) content.write(buildStsdBox(width, height, sps, pps))
content.write(buildEmptySttsBox()) content.write(buildSttsBox()) // Contains default timing for ffprobe frame rate detection
content.write(buildEmptyStscBox()) content.write(buildEmptyStscBox())
content.write(buildEmptyStszBox()) content.write(buildEmptyStszBox())
content.write(buildEmptyStcoBox()) content.write(buildEmptyStcoBox())
@@ -971,15 +753,29 @@ class HlsMuxer(
dos.writeShort(-1) // pre-defined dos.writeShort(-1) // pre-defined
output.write(buildAvcCBox(sps, pps)) output.write(buildAvcCBox(sps, pps))
output.write(buildPaspBox())
return wrapBox("avc1", output.toByteArray()) return wrapBox("avc1", output.toByteArray())
} }
/**
* Builds pixel aspect ratio box to explicitly declare square pixels (1:1).
* This helps players correctly interpret video dimensions without SAR scaling.
*/
private fun buildPaspBox(): ByteArray {
val output = ByteArrayOutputStream()
val dos = DataOutputStream(output)
dos.writeInt(1) // hSpacing (horizontal)
dos.writeInt(1) // vSpacing (vertical)
return wrapBox("pasp", output.toByteArray())
}
private fun buildAvcCBox(sps: ByteArray, pps: ByteArray): ByteArray { private fun buildAvcCBox(sps: ByteArray, pps: ByteArray): ByteArray {
val output = ByteArrayOutputStream() val output = ByteArrayOutputStream()
val dos = DataOutputStream(output) val dos = DataOutputStream(output)
// SPS layout: [0]=NAL header (0x67), [1]=profile_idc, [2]=constraint_flags, [3]=level_idc // SPS NAL unit format: [NAL header, profile_idc, constraint_flags, level_idc, ...]
// Skip byte 0 (NAL header, typically 0x67) to get the actual profile data
val profileIdc = if (sps.size > 1) sps[1].toInt() and 0xFF else 0x42 val profileIdc = if (sps.size > 1) sps[1].toInt() and 0xFF else 0x42
val profileCompat = if (sps.size > 2) sps[2].toInt() and 0xFF else 0x00 val profileCompat = if (sps.size > 2) sps[2].toInt() and 0xFF else 0x00
val levelIdc = if (sps.size > 3) sps[3].toInt() and 0xFF else 0x1F val levelIdc = if (sps.size > 3) sps[3].toInt() and 0xFF else 0x1F
@@ -1001,11 +797,21 @@ class HlsMuxer(
return wrapBox("avcC", output.toByteArray()) return wrapBox("avcC", output.toByteArray())
} }
private fun buildEmptySttsBox(): ByteArray { private fun buildSttsBox(): ByteArray {
val output = ByteArrayOutputStream() val output = ByteArrayOutputStream()
val dos = DataOutputStream(output) val dos = DataOutputStream(output)
// For fragmented MP4, stts is normally empty as timing is in trun boxes.
// However, ffprobe uses stts to calculate r_frame_rate when present.
// We add a single entry with the default sample duration so ffprobe
// can derive: r_frame_rate = timescale / sample_delta = 30000/1000 = 30
val defaultSampleDuration = timescale / fps
dos.writeInt(0) // version & flags dos.writeInt(0) // version & flags
dos.writeInt(0) // entry count dos.writeInt(1) // entry count (1 entry for default timing)
dos.writeInt(1) // sample_count (indicates this is the default duration)
dos.writeInt(defaultSampleDuration) // sample_delta in timescale units
return wrapBox("stts", output.toByteArray()) return wrapBox("stts", output.toByteArray())
} }
@@ -1042,9 +848,10 @@ class HlsMuxer(
val output = ByteArrayOutputStream() val output = ByteArrayOutputStream()
val dos = DataOutputStream(output) val dos = DataOutputStream(output)
// Default sample duration: timescale / fps // Calculate default sample duration so ffprobe can derive correct fps
// Since timescale = fps * 1000, duration = 1000 for any fps // fps = timescale / default_sample_duration
val defaultSampleDuration = 1000 // At 30fps with timescale=30000: duration=1000, ffprobe calculates 30000/1000=30
val defaultSampleDuration = timescale / fps
dos.writeInt(0) // version & flags dos.writeInt(0) // version & flags
dos.writeInt(1) // track ID dos.writeInt(1) // track ID
@@ -1093,9 +900,11 @@ class HlsMuxer(
): ByteArray { ): ByteArray {
// Calculate sizes to determine data offset // Calculate sizes to determine data offset
val mfhdBox = buildMfhdBox(sequenceNumber) val mfhdBox = buildMfhdBox(sequenceNumber)
val tfhdSize = 8 + 8 // box header + content (version/flags + track_id) // tfhd: 8 header + 4 version/flags + 4 track_id + 4 duration + 4 size + 4 flags = 28 bytes
val tfhdSize = 8 + 20
val tfdtSize = 8 + 12 // box header + version 1 content val tfdtSize = 8 + 12 // box header + version 1 content
val trunSize = 8 + 12 + (samples.size * 12) // header + fixed + per-sample (no composition offset) // trun: 8 header + 12 fixed + per-sample (size + flags only, no duration)
val trunSize = 8 + 12 + (samples.size * 8)
val trafSize = 8 + tfhdSize + tfdtSize + trunSize val trafSize = 8 + tfhdSize + tfdtSize + trunSize
val moofSize = 8 + mfhdBox.size + trafSize val moofSize = 8 + mfhdBox.size + trafSize
@@ -1130,9 +939,21 @@ class HlsMuxer(
val output = ByteArrayOutputStream() val output = ByteArrayOutputStream()
val dos = DataOutputStream(output) val dos = DataOutputStream(output)
// Flags: default-base-is-moof (0x020000) // Calculate default sample duration for this fragment
dos.writeInt(0x00020000) // This helps ffprobe calculate correct frame rate when reading via HLS
val defaultSampleDuration = timescale / fps // e.g., 30000/30 = 1000
// Match iOS AVFoundation's tfhd structure (28 bytes total)
// Flags: default-base-is-moof (0x020000) + default-sample-duration (0x000008)
// + default-sample-size (0x000010) + default-sample-flags (0x000020)
val flags = 0x00020000 or 0x000008 or 0x000010 or 0x000020
dos.writeInt(flags)
dos.writeInt(1) // track ID dos.writeInt(1) // track ID
dos.writeInt(defaultSampleDuration) // default sample duration in timescale units
dos.writeInt(0) // default sample size (0 = variable, specified in trun)
dos.writeInt(0x01010000) // default sample flags (non-keyframe, depends on others)
Log.d(TAG, "tfhd: default_sample_duration=$defaultSampleDuration (timescale=$timescale, fps=$fps)")
return wrapBox("tfhd", output.toByteArray()) return wrapBox("tfhd", output.toByteArray())
} }
@@ -1155,19 +976,17 @@ class HlsMuxer(
val output = ByteArrayOutputStream() val output = ByteArrayOutputStream()
val dos = DataOutputStream(output) val dos = DataOutputStream(output)
// Flags: data-offset + sample-duration + sample-size + sample-flags // Flags: data-offset + sample-size + sample-flags
val flags = 0x000001 or 0x000100 or 0x000200 or 0x000400 // NOTE: We intentionally OMIT sample-duration (0x000100) so ffprobe uses
// the default_sample_duration from tfhd instead of per-sample durations.
// This ensures consistent frame rate calculation via HLS.
val flags = 0x000001 or 0x000200 or 0x000400
dos.writeInt(flags) dos.writeInt(flags)
dos.writeInt(samples.size) dos.writeInt(samples.size)
dos.writeInt(dataOffset) dos.writeInt(dataOffset)
// Use constant duration based on configured fps for consistent frame rate
// This ensures ffprobe reports correct fps instead of calculating from variable timing
val constantDuration = timescale / configuredFps // e.g., 30000/30 = 1000 ticks
Log.d(TAG, "Writing ${samples.size} samples with constant duration=${constantDuration} ticks (${configuredFps}fps)")
for (sample in samples) { for (sample in samples) {
dos.writeInt(constantDuration) // No duration - using default from tfhd
dos.writeInt(sample.data.size) dos.writeInt(sample.data.size)
dos.writeInt(buildSampleFlags(sample.isKeyFrame)) dos.writeInt(buildSampleFlags(sample.isKeyFrame))
} }

View File

@@ -8,6 +8,7 @@ import com.facebook.common.statfs.StatFsHelper
import com.mrousavy.camera.extensions.getRecommendedBitRate import com.mrousavy.camera.extensions.getRecommendedBitRate
import com.mrousavy.camera.types.Orientation import com.mrousavy.camera.types.Orientation
import com.mrousavy.camera.types.RecordVideoOptions import com.mrousavy.camera.types.RecordVideoOptions
import com.mrousavy.camera.types.StreamSegmentType
import com.mrousavy.camera.utils.FileUtils import com.mrousavy.camera.utils.FileUtils
import java.io.File import java.io.File
import android.os.Environment import android.os.Environment
@@ -27,9 +28,7 @@ class RecordingSession(
private val filePath: String, private val filePath: String,
private val callback: (video: Video) -> Unit, private val callback: (video: Video) -> Unit,
private val onError: (error: CameraError) -> Unit, private val onError: (error: CameraError) -> Unit,
private val allCallbacks: CameraSession.Callback, private val allCallbacks: CameraSession.Callback
// Use FragmentedRecordingManager for HLS-compatible fMP4 output
private val useFragmentedMp4: Boolean = true
) { ) {
companion object { companion object {
private const val TAG = "RecordingSession" private const val TAG = "RecordingSession"
@@ -44,13 +43,15 @@ class RecordingSession(
data class Video(val path: String, val durationMs: Long, val size: Size) data class Video(val path: String, val durationMs: Long, val size: Size)
private val outputPath: File = File(filePath) // Normalize path - expo-file-system passes file:// URIs but File expects raw paths
// Handle both file:// and file:/ variants
private val outputPath: File = File(filePath.replace(Regex("^file:/+"), "/"))
private val bitRate = getBitRate() private val bitRate = getBitRate()
// Use FragmentedRecordingManager for HLS-compatible fMP4 output, // Use FragmentedRecordingManager for HLS-compatible fMP4 output,
// or fall back to ChunkedRecordingManager for regular MP4 chunks // or fall back to ChunkedRecordingManager for regular MP4 chunks
private val recorder: ChunkedRecorderInterface = if (useFragmentedMp4) { private val recorder: ChunkedRecorderInterface = if (options.streamSegmentType == StreamSegmentType.FRAGMENTED_MP4) {
FragmentedRecordingManager.fromParams( FragmentedRecordingManager.fromParams(
allCallbacks, allCallbacks,
size, size,
@@ -82,7 +83,7 @@ class RecordingSession(
fun start() { fun start() {
synchronized(this) { synchronized(this) {
Log.i(TAG, "Starting RecordingSession..") Log.i(TAG, "Starting RecordingSession with ${options.streamSegmentType} recorder..")
startTime = System.currentTimeMillis() startTime = System.currentTimeMillis()
recorder.start() recorder.start()
} }

View File

@@ -9,6 +9,7 @@ class RecordVideoOptions(map: ReadableMap) {
var videoBitRateOverride: Double? = null var videoBitRateOverride: Double? = null
var videoBitRateMultiplier: Double? = null var videoBitRateMultiplier: Double? = null
var orientation: Orientation? = null var orientation: Orientation? = null
var streamSegmentType: StreamSegmentType = StreamSegmentType.FRAGMENTED_MP4
init { init {
if (map.hasKey("fileType")) { if (map.hasKey("fileType")) {
@@ -29,5 +30,8 @@ class RecordVideoOptions(map: ReadableMap) {
if (map.hasKey("orientation")) { if (map.hasKey("orientation")) {
orientation = Orientation.fromUnionValue(map.getString("orientation")) orientation = Orientation.fromUnionValue(map.getString("orientation"))
} }
if (map.hasKey("streamSegmentType")) {
streamSegmentType = StreamSegmentType.fromUnionValue(map.getString("streamSegmentType"))
}
} }
} }

View File

@@ -0,0 +1,15 @@
package com.mrousavy.camera.types
enum class StreamSegmentType(override val unionValue: String) : JSUnionValue {
FRAGMENTED_MP4("FRAGMENTED_MP4"),
RB_CHUNKED_MP4("RB_CHUNKED_MP4");
companion object : JSUnionValue.Companion<StreamSegmentType> {
override fun fromUnionValue(unionValue: String?): StreamSegmentType =
when (unionValue) {
"FRAGMENTED_MP4" -> FRAGMENTED_MP4
"RB_CHUNKED_MP4" -> RB_CHUNKED_MP4
else -> FRAGMENTED_MP4 // Default to fMP4
}
}
}

View File

@@ -20,6 +20,9 @@ extension CameraSession {
onError: @escaping (_ error: CameraError) -> Void) { onError: @escaping (_ error: CameraError) -> Void) {
// Run on Camera Queue // Run on Camera Queue
CameraQueues.cameraQueue.async { CameraQueues.cameraQueue.async {
// Normalize path - expo-file-system passes file:// URIs but FileManager expects raw paths
let normalizedPath = filePath.hasPrefix("file://") ? String(filePath.dropFirst(7)) : filePath
let start = DispatchTime.now() let start = DispatchTime.now()
ReactLogger.log(level: .info, message: "Starting Video recording...") ReactLogger.log(level: .info, message: "Starting Video recording...")
@@ -38,11 +41,27 @@ extension CameraSession {
// Callback for when new chunks are ready // Callback for when new chunks are ready
let onChunkReady: (ChunkedRecorder.Chunk) -> Void = { chunk in let onChunkReady: (ChunkedRecorder.Chunk) -> Void = { chunk in
guard let delegate = self.delegate else { guard let delegate = self.delegate else {
ReactLogger.log(level: .warning, message: "Chunk ready but delegate is nil, dropping chunk: \(chunk)")
return return
} }
delegate.onVideoChunkReady(chunk: chunk) delegate.onVideoChunkReady(chunk: chunk)
} }
// Callback for when a chunk write fails (e.g. init file write failure)
let onChunkError: (Error) -> Void = { error in
ReactLogger.log(level: .error, message: "Chunk write error, stopping recording: \(error.localizedDescription)")
// Stop recording immediately
if let session = self.recordingSession {
session.stop(clock: self.captureSession.clock)
}
// Surface error to RN
if let cameraError = error as? CameraError {
onError(cameraError)
} else {
onError(.capture(.fileError))
}
}
// Callback for when the recording ends // Callback for when the recording ends
let onFinish = { (recordingSession: RecordingSession, status: AVAssetWriter.Status, error: Error?) in let onFinish = { (recordingSession: RecordingSession, status: AVAssetWriter.Status, error: Error?) in
defer { defer {
@@ -82,22 +101,23 @@ extension CameraSession {
} }
} }
if !FileManager.default.fileExists(atPath: filePath) { if !FileManager.default.fileExists(atPath: normalizedPath) {
do { do {
try FileManager.default.createDirectory(atPath: filePath, withIntermediateDirectories: true) try FileManager.default.createDirectory(atPath: normalizedPath, withIntermediateDirectories: true)
} catch { } catch {
onError(.capture(.createRecordingDirectoryError(message: error.localizedDescription))) onError(.capture(.createRecordingDirectoryError(message: error.localizedDescription)))
return return
} }
} }
ReactLogger.log(level: .info, message: "Will record to temporary file: \(filePath)") ReactLogger.log(level: .info, message: "Will record to temporary file: \(normalizedPath)")
do { do {
// Create RecordingSession for the temp file // Create RecordingSession for the temp file
let recordingSession = try RecordingSession(outputDiretory: filePath, let recordingSession = try RecordingSession(outputDiretory: normalizedPath,
fileType: options.fileType, fileType: options.fileType,
onChunkReady: onChunkReady, onChunkReady: onChunkReady,
onChunkError: onChunkError,
completion: onFinish) completion: onFinish)
// Init Audio + Activate Audio Session (optional) // Init Audio + Activate Audio Session (optional)

View File

@@ -24,12 +24,14 @@ class ChunkedRecorder: NSObject {
let outputURL: URL let outputURL: URL
let onChunkReady: ((Chunk) -> Void) let onChunkReady: ((Chunk) -> Void)
let onError: ((Error) -> Void)?
private var chunkIndex: UInt64 = 0 private var chunkIndex: UInt64 = 0
init(outputURL: URL, onChunkReady: @escaping ((Chunk) -> Void)) throws { init(outputURL: URL, onChunkReady: @escaping ((Chunk) -> Void), onError: ((Error) -> Void)? = nil) throws {
self.outputURL = outputURL self.outputURL = outputURL
self.onChunkReady = onChunkReady self.onChunkReady = onChunkReady
self.onError = onError
guard FileManager.default.fileExists(atPath: outputURL.path) else { guard FileManager.default.fileExists(atPath: outputURL.path) else {
throw CameraError.unknown(message: "output directory does not exist at: \(outputURL.path)", cause: nil) throw CameraError.unknown(message: "output directory does not exist at: \(outputURL.path)", cause: nil)
} }
@@ -56,28 +58,36 @@ extension ChunkedRecorder: AVAssetWriterDelegate {
private func saveInitSegment(_ data: Data) { private func saveInitSegment(_ data: Data) {
let url = outputURL.appendingPathComponent("init.mp4") let url = outputURL.appendingPathComponent("init.mp4")
save(data: data, url: url) do {
onChunkReady(url: url, type: .initialization) try data.write(to: url)
onChunkReady(url: url, type: .initialization)
} catch {
ReactLogger.log(level: .error, message: "Failed to write init file \(url): \(error.localizedDescription)")
onError?(CameraError.capture(.fileError))
}
} }
private func saveSegment(_ data: Data, report: AVAssetSegmentReport?) { private func saveSegment(_ data: Data, report: AVAssetSegmentReport?) {
let name = "\(chunkIndex).mp4" let name = "\(chunkIndex).mp4"
let url = outputURL.appendingPathComponent(name) let url = outputURL.appendingPathComponent(name)
save(data: data, url: url) if save(data: data, url: url) {
let duration = report? let duration = report?
.trackReports .trackReports
.filter { $0.mediaType == .video } .filter { $0.mediaType == .video }
.first? .first?
.duration .duration
onChunkReady(url: url, type: .data(index: chunkIndex, duration: duration)) onChunkReady(url: url, type: .data(index: chunkIndex, duration: duration))
chunkIndex += 1 chunkIndex += 1
}
} }
private func save(data: Data, url: URL) { private func save(data: Data, url: URL) -> Bool {
do { do {
try data.write(to: url) try data.write(to: url)
return true
} catch { } catch {
ReactLogger.log(level: .error, message: "Unable to write \(url): \(error.localizedDescription)") ReactLogger.log(level: .error, message: "Unable to write \(url): \(error.localizedDescription)")
return false
} }
} }

View File

@@ -74,12 +74,13 @@ class RecordingSession {
init(outputDiretory: String, init(outputDiretory: String,
fileType: AVFileType, fileType: AVFileType,
onChunkReady: @escaping ((ChunkedRecorder.Chunk) -> Void), onChunkReady: @escaping ((ChunkedRecorder.Chunk) -> Void),
onChunkError: ((Error) -> Void)? = nil,
completion: @escaping (RecordingSession, AVAssetWriter.Status, Error?) -> Void) throws { completion: @escaping (RecordingSession, AVAssetWriter.Status, Error?) -> Void) throws {
completionHandler = completion completionHandler = completion
do { do {
let outputURL = URL(fileURLWithPath: outputDiretory) let outputURL = URL(fileURLWithPath: outputDiretory)
recorder = try ChunkedRecorder(outputURL: outputURL, onChunkReady: onChunkReady) recorder = try ChunkedRecorder(outputURL: outputURL, onChunkReady: onChunkReady, onError: onChunkError)
assetWriter = AVAssetWriter(contentType: UTType(fileType.rawValue)!) assetWriter = AVAssetWriter(contentType: UTType(fileType.rawValue)!)
assetWriter.shouldOptimizeForNetworkUse = false assetWriter.shouldOptimizeForNetworkUse = false
assetWriter.outputFileTypeProfile = .mpeg4AppleHLS assetWriter.outputFileTypeProfile = .mpeg4AppleHLS

View File

@@ -41,6 +41,17 @@ export interface RecordVideoOptions {
* @default 'normal' * @default 'normal'
*/ */
videoBitRate?: 'extra-low' | 'low' | 'normal' | 'high' | 'extra-high' | number videoBitRate?: 'extra-low' | 'low' | 'normal' | 'high' | 'extra-high' | number
/**
* The stream segment type for recording on Android.
* - `FRAGMENTED_MP4`: HLS-compatible segments (init.mp4 + numbered segments)
* - `RB_CHUNKED_MP4`: Legacy chunked MP4 format
*
* iOS always uses FRAGMENTED_MP4 regardless of this setting.
*
* @platform android
* @default 'FRAGMENTED_MP4'
*/
streamSegmentType?: 'FRAGMENTED_MP4' | 'RB_CHUNKED_MP4'
} }
/** /**