diff --git a/gamdl/downloader/amdecrypt.py b/gamdl/downloader/amdecrypt.py
index 9649e16..9f7806f 100644
--- a/gamdl/downloader/amdecrypt.py
+++ b/gamdl/downloader/amdecrypt.py
@@ -168,10 +168,17 @@ def extract_song(input_path: str) -> SongInfo:
     trex_defaults = (
         _extract_trex_defaults(song_info.moov_data, audio_track_id)
         if song_info.moov_data
-        else {"default_sample_duration": 1024, "default_sample_size": 0}
+        else None
     )
-    default_sample_duration = trex_defaults["default_sample_duration"]
-    default_sample_size = trex_defaults["default_sample_size"]
+    if trex_defaults:
+        default_sample_duration = trex_defaults["default_sample_duration"]
+        default_sample_size = trex_defaults["default_sample_size"]
+    else:
+        # Fallback defaults. ALAC typically uses 4096 samples per frame,
+        # while AAC uses 1024. Default to 4096 if the track contains 'alac'.
+        is_alac = song_info.moov_data and b"alac" in song_info.moov_data
+        default_sample_duration = 4096 if is_alac else 1024
+        default_sample_size = 0
     logger.debug(
         f"Default sample duration: {default_sample_duration}, "
         f"default sample size: {default_sample_size}"
@@ -210,6 +217,18 @@ def extract_song(input_path: str) -> SongInfo:
             song_info.samples.extend(samples_from_pair)
             moof_box = None
 
+    # Post-process samples: if this is ALAC, ensure all samples have duration 4096.
+    # Apple Music fragments often report 1024 in trex/tfhd defaults, but 
+    # ALAC frames are actually 4096 samples long. This mismatch is the 
+    # root cause of the 1:16 duration reporting for 5-minute tracks.
+    is_alac = song_info.moov_data and (b"alac" in song_info.moov_data or b"ALAC" in song_info.moov_data)
+    if is_alac:
+        logger.debug("ALAC detected: forcing all sample durations to 4096")
+        for sample in song_info.samples:
+            # Only override if it was 0 or the common incorrect default of 1024
+            if sample.duration in (0, 1024):
+                sample.duration = 4096
+
     logger.debug(f"Extracted {len(song_info.samples)} samples from {input_path}")
     return song_info
 
@@ -596,10 +615,12 @@ def write_decrypted_m4a(
     orig_mvhd = None
     orig_tkhd = None
     orig_mdhd = None
-    orig_hdlr = None
     orig_smhd = None
     orig_dinf = None
-    timescale = 44100  # Default
+    # We will use the actual audio sample rate from the stsd as our 
+    # master timescale to ensure 100% duration consistency.
+    orig_hdlr = None
+    timescale = 44100  # Default fallback
 
     if original_path:
         with open(original_path, "rb") as f:
@@ -611,7 +632,8 @@ def write_decrypted_m4a(
 
     if orig_data:
         stsd_content = _extract_stsd_content(orig_data)
-        timescale = _extract_timescale(orig_data)
+        # Extract the REAL sample rate from the codec configuration
+        timescale = _extract_sample_rate_from_stsd(stsd_content) or _extract_timescale(orig_data)
 
         # Find moov box and extract child boxes
         moov_idx = orig_data.find(b"moov")
@@ -712,7 +734,7 @@ def _write_moov(
 
     # mvhd (movie header)
     if orig_mvhd:
-        f.write(_patch_mvhd_duration(orig_mvhd, total_duration))
+        f.write(_patch_mvhd_duration(orig_mvhd, total_duration, timescale))
     else:
         mvhd_content = struct.pack(">II", 0, 0)  # creation, modification
         mvhd_content += struct.pack(">I", timescale)
@@ -755,7 +777,7 @@ def _write_moov(
 
     # mdhd (media header) - preserves original language code
     if orig_mdhd:
-        f.write(_patch_mdhd_duration(orig_mdhd, total_duration))
+        f.write(_patch_mdhd_duration(orig_mdhd, total_duration, timescale))
     else:
         mdhd_content = struct.pack(">II", 0, 0)  # creation, modification
         mdhd_content += struct.pack(">I", timescale)
@@ -845,6 +867,23 @@ def _write_moov(
     f.write(struct.pack(">I", mdat_offset))
     f.seek(0, 2)  # Back to end
 
+def _extract_sample_rate_from_stsd(stsd_content: bytes) -> Optional[int]:
+    """Extract the actual audio sample rate from the stsd box content."""
+    # Header: version(1)+flags(3)+count(4) + Entry: size(4)+type(4) = 16 bytes
+    # AudioSampleEntry v0: reserved(6)+dref(2)+ver(2)+rev(2)+vend(4)+chan(2)+size(2)+comp(2)+pack(2)+rate(4)
+    # The fixed-point sample_rate field is at offset 16 + 24 = 40.
+    if not stsd_content or len(stsd_content) < 44:
+        return None
+        
+    samplerate_offset = 40
+    sample_rate_fixed = struct.unpack(">I", stsd_content[samplerate_offset : samplerate_offset + 4])[0]
+    sample_rate = sample_rate_fixed >> 16
+    
+    # Sanity check: standard audio sample rates should be between 8000 and 384000
+    if 8000 <= sample_rate <= 384000:
+        return sample_rate
+    return None
+
 
 def _write_stsd(f, stsd_content: bytes):
     """Write sample description box using content from original file.
@@ -1188,10 +1227,16 @@ def _extract_timescale(data: bytes) -> int:
     """Extract timescale from moov/mvhd or mdhd box."""
     # Look for mdhd box (media header has the audio timescale)
     idx = data.find(b"mdhd")
-    if idx > 0 and idx + 24 < len(data):
-        # mdhd: version(1) + flags(3) + creation(4) + modification(4) + timescale(4)
-        return struct.unpack(">I", data[idx + 16 : idx + 20])[0]
-    return 44100  # Default
+    if idx > 0 and idx + 28 < len(data):
+        # mdhd: size(4) + type(b'mdhd') + version(1) + flags(3)
+        version = data[idx + 4]
+        if version == 0:
+            # v0: ver+flags(4) + creation(4) + modification(4) + timescale(4)
+            return struct.unpack(">I", data[idx + 16 : idx + 20])[0]
+        else:
+            # v1: ver+flags(4) + creation(8) + modification(8) + timescale(4)
+            return struct.unpack(">I", data[idx + 24 : idx + 28])[0]
+    return 44100  # Default fallback
 
 
 def _find_child_box(
@@ -1244,15 +1289,17 @@ def _find_audio_trak(moov_data: bytes) -> Optional[bytes]:
     return None
 
 
-def _patch_mvhd_duration(box_data: bytes, duration: int) -> bytes:
-    """Return a copy of the mvhd box with its duration field patched."""
+def _patch_mvhd_duration(box_data: bytes, duration: int, timescale: int) -> bytes:
+    """Return a copy of the mvhd box with its duration and timescale fields patched."""
     data = bytearray(box_data)
     version = data[8]  # After size(4) + type(4)
     if version == 0:
         # v0: ver+flags(4) + creation(4) + modification(4) + timescale(4) + duration(4)
+        struct.pack_into(">I", data, 20, timescale)
         struct.pack_into(">I", data, 24, duration)
     else:
         # v1: ver+flags(4) + creation(8) + modification(8) + timescale(4) + duration(8)
+        struct.pack_into(">I", data, 28, timescale)
         struct.pack_into(">Q", data, 32, duration)
     return bytes(data)
 
@@ -1272,17 +1319,18 @@ def _patch_tkhd_duration(box_data: bytes, duration: int) -> bytes:
     return bytes(data)
 
 
-def _patch_mdhd_duration(box_data: bytes, duration: int) -> bytes:
-    """Return a copy of the mdhd box with its duration field patched.
 
-    Preserves the original language code and all other fields.
-    """
+def _patch_mdhd_duration(box_data: bytes, duration: int, timescale: int) -> bytes:
+    """Return a copy of the mdhd box with its duration and timescale fields patched."""
     data = bytearray(box_data)
     version = data[8]
     if version == 0:
-        # Same layout as mvhd v0
+        # v0: ver+flags(4) + creation(4) + modification(4) + timescale(4) + duration(4)
+        struct.pack_into(">I", data, 20, timescale)
         struct.pack_into(">I", data, 24, duration)
     else:
+        # v1: ver+flags(4) + creation(8) + modification(8) + timescale(4) + duration(8)
+        struct.pack_into(">I", data, 28, timescale)
         struct.pack_into(">Q", data, 32, duration)
     return bytes(data)
 
@@ -1331,8 +1379,13 @@ def _extract_trex_defaults(moov_data: bytes, target_track_id: int = 0) -> dict:
         Dict with keys: default_sample_duration, default_sample_size,
         default_sample_description_index, default_sample_flags.
     """
+    # Determine fallback duration based on codec
+    # ALAC frames are 4096 samples, AAC frames are 1024 samples
+    is_alac = b"alac" in moov_data or b"ALAC" in moov_data
+    fallback_duration = 4096 if is_alac else 1024
+
     defaults = {
-        "default_sample_duration": 1024,
+        "default_sample_duration": fallback_duration,
         "default_sample_size": 0,
         "default_sample_description_index": 1,
         "default_sample_flags": 0,
@@ -1361,12 +1414,17 @@ def _extract_trex_defaults(moov_data: bytes, target_track_id: int = 0) -> dict:
                 defaults["default_sample_description_index"] = struct.unpack(
                     ">I", trex_data[16:20]
                 )[0]
-                defaults["default_sample_duration"] = struct.unpack(
-                    ">I", trex_data[20:24]
-                )[0]
-                defaults["default_sample_size"] = struct.unpack(">I", trex_data[24:28])[
-                    0
-                ]
+                
+                # Extract duration and protect against Apple's dummy values
+                parsed_duration = struct.unpack(">I", trex_data[20:24])[0]
+                
+                # Override if the provider wrote 0, or if they incorrectly wrote 1024 for an ALAC track
+                if parsed_duration == 0 or (is_alac and parsed_duration == 1024):
+                    defaults["default_sample_duration"] = fallback_duration
+                else:
+                    defaults["default_sample_duration"] = parsed_duration
+
+                defaults["default_sample_size"] = struct.unpack(">I", trex_data[24:28])[0]
                 defaults["default_sample_flags"] = struct.unpack(
                     ">I", trex_data[28:32]
                 )[0]
@@ -1380,7 +1438,6 @@ def _extract_trex_defaults(moov_data: bytes, target_track_id: int = 0) -> dict:
 
     return defaults
 
-
 def _extract_encryption_info(moov_data: bytes) -> Optional[EncryptionInfo]:
     """Extract encryption scheme info from the audio track's sinf box.