diff --git a/src/Autoload/Export.gd b/src/Autoload/Export.gd
index b60545e0b..ca839dcd6 100644
--- a/src/Autoload/Export.gd
+++ b/src/Autoload/Export.gd
@@ -536,15 +536,15 @@ func export_video(export_paths: PackedStringArray, project: Project) -> bool:
 	var audio_layer_count := 0
 	var max_audio_duration := 0
 	var adelay_string := ""
-	for layer in project.layers:
-		if layer is AudioLayer and is_instance_valid(layer.audio) and layer.audio is AudioStreamMP3:
+	for layer in project.get_all_audio_layers():
+		if layer.audio is AudioStreamMP3:
 			var temp_file_name := str(audio_layer_count + 1).pad_zeros(number_of_digits) + ".mp3"
 			var temp_file_path := temp_path_real.path_join(temp_file_name)
 			var temp_audio_file := FileAccess.open(temp_file_path, FileAccess.WRITE)
 			temp_audio_file.store_buffer(layer.audio.data)
 			ffmpeg_combine_audio.append("-i")
 			ffmpeg_combine_audio.append(temp_file_path)
-			var delay: int = layer.playback_position * 1000
+			var delay := floori(layer.playback_position * 1000)
 			# [n]adelay=delay_in_ms:all=1[na]
 			adelay_string += (
 				"[%s]adelay=%s:all=1[%sa];" % [audio_layer_count, delay, audio_layer_count]
diff --git a/src/Autoload/OpenSave.gd b/src/Autoload/OpenSave.gd
index d8fac67fe..868d25ee1 100644
--- a/src/Autoload/OpenSave.gd
+++ b/src/Autoload/OpenSave.gd
@@ -201,7 +201,7 @@ func handle_loading_video(file: String) -> bool:
 	var output_audio_file := temp_path_real.path_join("audio.mp3")
 	# ffmpeg -y -i input_file -vn audio.mp3
 	var ffmpeg_execute_audio: PackedStringArray = ["-y", "-i", file, "-vn", output_audio_file]
-	var success_audio := OS.execute(Global.ffmpeg_path, ffmpeg_execute_audio, [], true)
+	OS.execute(Global.ffmpeg_path, ffmpeg_execute_audio, [], true)
 	if FileAccess.file_exists(output_audio_file):
 		open_audio_file(output_audio_file)
 		temp_dir.remove("audio.mp3")
@@ -448,6 +448,14 @@ func save_pxo_file(
 			zip_packer.start_file(tileset_path.path_join(str(j)))
 			zip_packer.write_file(tile.image.get_data())
 			zip_packer.close_file()
+	var audio_layers := project.get_all_audio_layers()
+	for i in audio_layers.size():
+		var layer := audio_layers[i]
+		var audio_path := "audio/%s" % i
+		if layer.audio is AudioStreamMP3:
+			zip_packer.start_file(audio_path)
+			zip_packer.write_file(layer.audio.data)
+			zip_packer.close_file()
 	zip_packer.close()
 
 	if temp_path != path:
@@ -914,7 +922,6 @@ func set_new_imported_tab(project: Project, path: String) -> void:
 
 func open_audio_file(path: String) -> void:
 	var audio_stream: AudioStream
-	var file_ext := path.get_extension().to_lower()
 	var file := FileAccess.open(path, FileAccess.READ)
 	audio_stream = AudioStreamMP3.new()
 	audio_stream.data = file.get_buffer(file.get_length())
diff --git a/src/Classes/Layers/AudioLayer.gd b/src/Classes/Layers/AudioLayer.gd
index b1da55b0a..2bf49607d 100644
--- a/src/Classes/Layers/AudioLayer.gd
+++ b/src/Classes/Layers/AudioLayer.gd
@@ -26,14 +26,26 @@ func get_audio_length() -> float:
 		return -1.0
 
 
+func get_audio_type() -> String:
+	if not is_instance_valid(audio):
+		return ""
+	return audio.get_class()
+
+
 # Overridden Methods:
 func serialize() -> Dictionary:
-	var data := {"name": name, "type": get_layer_type()}
+	var data := {
+		"name": name,
+		"type": get_layer_type(),
+		"playback_frame": playback_frame,
+		"audio_type": get_audio_type()
+	}
 	return data
 
 
 func deserialize(dict: Dictionary) -> void:
 	super.deserialize(dict)
+	playback_frame = dict.get("playback_frame", playback_frame)
 
 
 func get_layer_type() -> int:
diff --git a/src/Classes/Project.gd b/src/Classes/Project.gd
index 438401031..fcec8ee2a 100644
--- a/src/Classes/Project.gd
+++ b/src/Classes/Project.gd
@@ -360,6 +360,7 @@ func deserialize(dict: Dictionary, zip_reader: ZIPReader = null, file: FileAcces
 			tileset.deserialize(saved_tileset)
 			tilesets.append(tileset)
 	if dict.has("frames") and dict.has("layers"):
+		var audio_layers := 0
 		for saved_layer in dict.layers:
 			match int(saved_layer.get("type", Global.LayerTypes.PIXEL)):
 				Global.LayerTypes.PIXEL:
@@ -370,6 +371,18 @@ func deserialize(dict: Dictionary, zip_reader: ZIPReader = null, file: FileAcces
 					layers.append(Layer3D.new(self))
 				Global.LayerTypes.TILEMAP:
 					layers.append(LayerTileMap.new(self, null))
+				Global.LayerTypes.AUDIO:
+					var layer := AudioLayer.new(self)
+					var audio_path := "audio/%s" % audio_layers
+					if zip_reader.file_exists(audio_path):
+						var audio_data := zip_reader.read_file(audio_path)
+						var stream: AudioStream
+						if saved_layer.get("audio_type", "") == "AudioStreamMP3":
+							stream = AudioStreamMP3.new()
+							stream.data = audio_data
+						layer.audio = stream
+					layers.append(layer)
+					audio_layers += 1
 
 		var frame_i := 0
 		for frame in dict.frames:
@@ -394,6 +407,8 @@ func deserialize(dict: Dictionary, zip_reader: ZIPReader = null, file: FileAcces
 						var tileset := tilesets[tileset_index]
 						var new_cel := CelTileMap.new(tileset, image)
 						cels.append(new_cel)
+					Global.LayerTypes.AUDIO:
+						cels.append(AudioCel.new())
 				cel["pxo_version"] = pxo_version
 				cels[cel_i].deserialize(cel)
 				_deserialize_metadata(cels[cel_i], cel)
@@ -662,6 +677,18 @@ func get_all_pixel_cels() -> Array[PixelCel]:
 	return cels
 
 
+func get_all_audio_layers(only_valid_streams := true) -> Array[AudioLayer]:
+	var audio_layers: Array[AudioLayer]
+	for layer in layers:
+		if layer is AudioLayer:
+			if only_valid_streams:
+				if is_instance_valid(layer.audio):
+					audio_layers.append(layer)
+			else:
+				audio_layers.append(layer)
+	return audio_layers
+
+
 ## Reads data from [param cels] and appends them to [param data],
 ## to be used for the undo/redo system.
 ## It adds data such as the images of [PixelCel]s,