auto-editor 24.27.1__py3-none-any.whl → 24.30.1__py3-none-any.whl

auto_editor/__init__.py

@@ -1,2 +1,2 @@
-__version__ = "24.27.1"
-version = "24w27a"
+__version__ = "24.30.1"
+version = "24w30a"

auto_editor/analyze.py

@@ -4,76 +4,32 @@ import os
 import re
 from dataclasses import dataclass
 from fractions import Fraction
+from math import ceil
 from typing import TYPE_CHECKING
 
+import av
 import numpy as np
+from av.audio.fifo import AudioFifo
+from av.subtitles.subtitle import AssSubtitle
 
 from auto_editor import version
-from auto_editor.lang.json import Lexer, Parser, dump
-from auto_editor.lib.contracts import (
-    is_bool,
-    is_nat,
-    is_nat1,
-    is_str,
-    is_threshold,
-    is_void,
-    orc,
-)
-from auto_editor.lib.data_structs import Sym
-from auto_editor.utils.cmdkw import (
-    Required,
-    pAttr,
-    pAttrs,
-)
 from auto_editor.utils.subtitle_tools import convert_ass_to_text
-from auto_editor.wavfile import read
 
 if TYPE_CHECKING:
+    from collections.abc import Iterator
     from fractions import Fraction
     from typing import Any
 
-    from av.filter import FilterContext
     from numpy.typing import NDArray
 
     from auto_editor.ffwrapper import FileInfo
-    from auto_editor.output import Ensure
     from auto_editor.utils.bar import Bar
     from auto_editor.utils.log import Log
 
 
-audio_builder = pAttrs(
-    "audio",
-    pAttr("threshold", 0.04, is_threshold),
-    pAttr("stream", 0, orc(is_nat, Sym("all"), "all")),
-    pAttr("mincut", 6, is_nat),
-    pAttr("minclip", 3, is_nat),
-)
-motion_builder = pAttrs(
-    "motion",
-    pAttr("threshold", 0.02, is_threshold),
-    pAttr("stream", 0, is_nat),
-    pAttr("blur", 9, is_nat),
-    pAttr("width", 400, is_nat1),
-)
-subtitle_builder = pAttrs(
-    "subtitle",
-    pAttr("pattern", Required, is_str),
-    pAttr("stream", 0, is_nat),
-    pAttr("ignore-case", False, is_bool),
-    pAttr("max-count", None, orc(is_nat, is_void)),
-)
-
-builder_map = {
-    "audio": audio_builder,
-    "motion": motion_builder,
-    "subtitle": subtitle_builder,
-}
-
-
 @dataclass(slots=True)
 class FileSetup:
     src: FileInfo
-    ensure: Ensure
     strict: bool
     tb: Fraction
     bar: Bar
@@ -85,15 +41,6 @@ class LevelError(Exception):
     pass
 
 
-def link_nodes(*nodes: FilterContext) -> None:
-    for c, n in zip(nodes, nodes[1:]):
-        c.link_to(n)
-
-
-def to_threshold(arr: np.ndarray, t: int | float) -> NDArray[np.bool_]:
-    return np.fromiter((x >= t for x in arr), dtype=np.bool_)
-
-
 def mut_remove_small(
     arr: NDArray[np.bool_], lim: int, replace: int, with_: int
 ) -> None:
@@ -141,9 +88,90 @@ def obj_tag(tag: str, tb: Fraction, obj: dict[str, Any]) -> str:
     return key
 
 
+def iter_audio(src, tb: Fraction, stream: int = 0) -> Iterator[float]:
+    fifo = AudioFifo()
+    try:
+        container = av.open(src.path, "r")
+        audio_stream = container.streams.audio[stream]
+        sample_rate = audio_stream.rate
+
+        exact_size = (1 / tb) * sample_rate
+        accumulated_error = 0
+
+        # Resample so that audio data is between [-1, 1]
+        resampler = av.AudioResampler(
+            av.AudioFormat("flt"), audio_stream.layout, sample_rate
+        )
+
+        for frame in container.decode(audio=stream):
+            frame.pts = None  # Skip time checks
+
+            for reframe in resampler.resample(frame):
+                fifo.write(reframe)
+
+            while fifo.samples >= ceil(exact_size):
+                size_with_error = exact_size + accumulated_error
+                current_size = round(size_with_error)
+                accumulated_error = size_with_error - current_size
+
+                audio_chunk = fifo.read(current_size)
+                assert audio_chunk is not None
+                arr = audio_chunk.to_ndarray().flatten()
+                yield float(np.max(np.abs(arr)))
+
+    finally:
+        container.close()
+
+
+def iter_motion(src, tb, stream: int, blur: int, width: int) -> Iterator[float]:
+    container = av.open(src.path, "r")
+
+    video = container.streams.video[stream]
+    video.thread_type = "AUTO"
+
+    prev_frame = None
+    current_frame = None
+    total_pixels = src.videos[0].width * src.videos[0].height
+    index = 0
+    prev_index = -1
+
+    graph = av.filter.Graph()
+    graph.link_nodes(
+        graph.add_buffer(template=video),
+        graph.add("scale", f"{width}:-1"),
+        graph.add("format", "gray"),
+        graph.add("gblur", f"sigma={blur}"),
+        graph.add("buffersink"),
+    ).configure()
+
+    for unframe in container.decode(video):
+        if unframe.pts is None:
+            continue
+
+        graph.push(unframe)
+        frame = graph.pull()
+        assert frame.time is not None
+        index = round(frame.time * tb)
+
+        current_frame = frame.to_ndarray()
+        if prev_frame is None:
+            value = 0.0
+        else:
+            # Use `int16` to avoid underflow with `uint8` datatype
+            diff = np.abs(prev_frame.astype(np.int16) - current_frame.astype(np.int16))
+            value = np.count_nonzero(diff) / total_pixels
+
+        for _ in range(index - prev_index):
+            yield value
+
+        prev_frame = current_frame
+        prev_index = index
+
+    container.close()
+
+
 @dataclass(slots=True)
 class Levels:
-    ensure: Ensure
     src: FileInfo
     tb: Fraction
     bar: Bar
@@ -156,26 +184,16 @@ class Levels:
             if (arr := self.read_cache("audio", {"stream": 0})) is not None:
                 return len(arr)
 
-            sr, samples = read(self.ensure.audio(self.src, 0))
-            samp_count = len(samples)
-            del samples
-
-            samp_per_ticks = sr / self.tb
-            ticks = int(samp_count / samp_per_ticks)
-            self.log.debug(f"Audio Length: {ticks}")
-            self.log.debug(
-                f"... without rounding: {float(samp_count / samp_per_ticks)}"
-            )
-            return ticks
+            result = sum(1 for _ in iter_audio(self.src, self.tb, 0))
+            self.log.debug(f"Audio Length: {result}")
+            return result
 
         # If there's no audio, get length in video metadata.
-        import av
-
-        with av.open(f"{self.src.path}") as cn:
-            if len(cn.streams.video) < 1:
+        with av.open(self.src.path) as container:
+            if len(container.streams.video) == 0:
                 self.log.error("Could not get media duration")
 
-            video = cn.streams.video[0]
+            video = container.streams.video[0]
 
             if video.duration is None or video.time_base is None:
                 dur = 0
@@ -193,107 +211,101 @@ class Levels:
 
     def read_cache(self, tag: str, obj: dict[str, Any]) -> None | np.ndarray:
         workfile = os.path.join(
-            os.path.dirname(self.temp), f"ae-{version}", "cache.json"
+            os.path.dirname(self.temp), f"ae-{version}", "cache.npz"
         )
 
         try:
-            with open(workfile, encoding="utf-8") as file:
-                cache = Parser(Lexer(workfile, file)).expr()
-        except Exception:
+            npzfile = np.load(workfile, allow_pickle=False)
+        except Exception as e:
+            self.log.debug(e)
             return None
 
-        if f"{self.src.path.resolve()}" not in cache:
+        key = f"{self.src.path}:{obj_tag(tag, self.tb, obj)}"
+        if key not in npzfile.files:
             return None
 
-        key = obj_tag(tag, self.tb, obj)
-
-        if key not in (root := cache[f"{self.src.path.resolve()}"]):
-            return None
-
-        return np.asarray(root[key]["arr"], dtype=root[key]["type"])
+        self.log.debug("Using cache")
+        return npzfile[key]
 
     def cache(self, tag: str, obj: dict[str, Any], arr: np.ndarray) -> np.ndarray:
         workdur = os.path.join(os.path.dirname(self.temp), f"ae-{version}")
-        workfile = os.path.join(workdur, "cache.json")
         if not os.path.exists(workdur):
             os.mkdir(workdur)
 
-        key = obj_tag(tag, self.tb, obj)
-
-        try:
-            with open(workfile, encoding="utf-8") as file:
-                json_object = Parser(Lexer(workfile, file)).expr()
-        except Exception:
-            json_object = {}
-
-        entry = {"type": str(arr.dtype), "arr": arr.tolist()}
-        src_key = f"{self.src.path}"
-
-        if src_key in json_object:
-            json_object[src_key][key] = entry
-        else:
-            json_object[src_key] = {key: entry}
-
-        with open(os.path.join(workdur, "cache.json"), "w", encoding="utf-8") as file:
-            dump(json_object, file)
+        tag = obj_tag(tag, self.tb, obj)
+        np.savez(os.path.join(workdur, "cache.npz"), **{f"{self.src.path}:{tag}": arr})
 
         return arr
 
-    def audio(self, s: int) -> NDArray[np.float64]:
-        if s > len(self.src.audios) - 1:
-            raise LevelError(f"audio: audio stream '{s}' does not exist.")
+    def audio(self, stream: int) -> NDArray[np.float64]:
+        if stream >= len(self.src.audios):
+            raise LevelError(f"audio: audio stream '{stream}' does not exist.")
 
-        if (arr := self.read_cache("audio", {"stream": s})) is not None:
+        if (arr := self.read_cache("audio", {"stream": stream})) is not None:
             return arr
 
-        sr, samples = read(self.ensure.audio(self.src, s))
-
-        if len(samples) == 0:
-            raise LevelError(f"audio: stream '{s}' has no samples.")
-
-        def get_max_volume(s: np.ndarray) -> float:
-            return max(float(np.max(s)), -float(np.min(s)))
-
-        max_volume = get_max_volume(samples)
-        self.log.debug(f"Max volume: {max_volume}")
+        with av.open(self.src.path, "r") as container:
+            audio = container.streams.audio[stream]
+            if audio.duration is not None and audio.time_base is not None:
+                inaccurate_dur = int(audio.duration * audio.time_base * self.tb)
+            elif container.duration is not None:
+                inaccurate_dur = int(container.duration / av.time_base * self.tb)
+            else:
+                inaccurate_dur = 1024
 
-        samp_count = samples.shape[0]
-        samp_per_ticks = sr / self.tb
+        bar = self.bar
+        bar.start(inaccurate_dur, "Analyzing audio volume")
 
-        if samp_per_ticks < 1:
-            self.log.error(
-                f"audio: stream '{s}'\n  Samplerate ({sr}) must be greater than "
-                f"or equal to timebase ({self.tb})\n"
-                "  Try `-fps 30` and/or `--sample-rate 48000`"
-            )
+        result = np.zeros((inaccurate_dur), dtype=np.float64)
+        index = 0
+        for value in iter_audio(self.src, self.tb, stream):
+            if index > len(result) - 1:
+                result = np.concatenate(
+                    (result, np.zeros((len(result)), dtype=np.float64))
+                )
+            result[index] = value
+            bar.tick(index)
+            index += 1
 
-        audio_ticks = int(samp_count / samp_per_ticks)
-        self.log.debug(
-            f"analyze: audio length: {audio_ticks} ({float(samp_count / samp_per_ticks)})"
-        )
-        self.bar.start(audio_ticks, "Analyzing audio volume")
+        bar.end()
+        return self.cache("audio", {"stream": stream}, result[:index])
 
-        threshold_list = np.zeros((audio_ticks), dtype=np.float64)
+    def motion(self, stream: int, blur: int, width: int) -> NDArray[np.float64]:
+        if stream >= len(self.src.videos):
+            raise LevelError(f"motion: video stream '{stream}' does not exist.")
 
-        if max_volume == 0:  # Prevent dividing by zero
-            return threshold_list
+        mobj = {"stream": stream, "width": width, "blur": blur}
+        if (arr := self.read_cache("motion", mobj)) is not None:
+            return arr
 
-        # Determine when audio is silent or loud.
-        for i in range(audio_ticks):
-            if i % 500 == 0:
-                self.bar.tick(i)
+        with av.open(self.src.path, "r") as container:
+            video = container.streams.video[stream]
+            inaccurate_dur = (
+                1024
+                if video.duration is None or video.time_base is None
+                else int(video.duration * video.time_base * self.tb)
+            )
 
-            start = int(i * samp_per_ticks)
-            end = min(int((i + 1) * samp_per_ticks), samp_count)
+        bar = self.bar
+        bar.start(inaccurate_dur, "Analyzing motion")
 
-            threshold_list[i] = get_max_volume(samples[start:end]) / max_volume
+        result = np.zeros((inaccurate_dur), dtype=np.float64)
+        index = 0
+        for value in iter_motion(self.src, self.tb, stream, blur, width):
+            if index > len(result) - 1:
+                result = np.concatenate(
+                    (result, np.zeros((len(result)), dtype=np.float64))
+                )
+            result[index] = value
+            bar.tick(index)
+            index += 1
 
-        self.bar.end()
-        return self.cache("audio", {"stream": s}, threshold_list)
+        bar.end()
+        return self.cache("motion", mobj, result[:index])
 
     def subtitle(
         self,
-        patterns: str,
+        pattern: str,
         stream: int,
         ignore_case: bool,
         max_count: int | None,
@@ -303,14 +315,10 @@ class Levels:
 
         try:
             flags = re.IGNORECASE if ignore_case else 0
-            pattern = re.compile(patterns, flags)
-            del patterns  # make sure we don't accidentally use it
+            re_pattern = re.compile(pattern, flags)
         except re.error as e:
             self.log.error(e)
 
-        import av
-        from av.subtitles.subtitle import AssSubtitle, TextSubtitle
-
         try:
             container = av.open(self.src.path, "r")
             subtitle_stream = container.streams.subtitles[stream]
@@ -357,80 +365,13 @@ class Levels:
                 for sub in subset:
                     if isinstance(sub, AssSubtitle):
                         line = convert_ass_to_text(sub.ass.decode(errors="ignore"))
-                    elif isinstance(sub, TextSubtitle):
-                        line = sub.text.decode(errors="ignore")
                     else:
                         continue
 
-                    if line and re.search(pattern, line):
+                    if line and re.search(re_pattern, line):
                         result[san_start:san_end] = 1
                         count += 1
 
         container.close()
 
         return result
-
-    def motion(self, s: int, blur: int, width: int) -> NDArray[np.float64]:
-        import av
-
-        if s >= len(self.src.videos):
-            raise LevelError(f"motion: video stream '{s}' does not exist.")
-
-        mobj = {"stream": s, "width": width, "blur": blur}
-        if (arr := self.read_cache("motion", mobj)) is not None:
-            return arr
-
-        container = av.open(f"{self.src.path}", "r")
-
-        stream = container.streams.video[s]
-        stream.thread_type = "AUTO"
-
-        inaccurate_dur = 1 if stream.duration is None else stream.duration
-        self.bar.start(inaccurate_dur, "Analyzing motion")
-
-        prev_frame = None
-        current_frame = None
-        total_pixels = self.src.videos[0].width * self.src.videos[0].height
-        index = 0
-
-        graph = av.filter.Graph()
-        link_nodes(
-            graph.add_buffer(template=stream),
-            graph.add("scale", f"{width}:-1"),
-            graph.add("format", "gray"),
-            graph.add("gblur", f"sigma={blur}"),
-            graph.add("buffersink"),
-        )
-        graph.configure()
-
-        threshold_list = np.zeros((1024), dtype=np.float64)
-
-        for unframe in container.decode(stream):
-            graph.push(unframe)
-            frame = graph.pull()
-
-            # Showing progress ...
-            assert frame.time is not None
-            index = int(frame.time * self.tb)
-            if frame.pts is not None:
-                self.bar.tick(frame.pts)
-
-            current_frame = frame.to_ndarray()
-
-            if index > len(threshold_list) - 1:
-                threshold_list = np.concatenate(
-                    (threshold_list, np.zeros((len(threshold_list)), dtype=np.float64)),
-                    axis=0,
-                )
-
-            if prev_frame is not None:
-                # Use `int16` to avoid underflow with `uint8` datatype
-                diff = np.abs(
-                    prev_frame.astype(np.int16) - current_frame.astype(np.int16)
-                )
-                threshold_list[index] = np.count_nonzero(diff) / total_pixels
-
-            prev_frame = current_frame
-
-        self.bar.end()
-        return self.cache("motion", mobj, threshold_list[:index])

auto_editor/edit.py

@@ -68,12 +68,8 @@ def set_video_codec(
 ) -> str:
     if codec == "auto":
         codec = "h264" if (src is None or not src.videos) else src.videos[0].codec
-        if ctr.vcodecs is not None:
-            if ctr.vstrict and codec not in ctr.vcodecs:
-                return ctr.vcodecs[0]
-
-            if codec in ctr.disallow_v:
-                return ctr.vcodecs[0]
+        if codec not in ctr.vcodecs and ctr.default_vid != "none":
+            return ctr.default_vid
         return codec
 
     if codec == "copy":
@@ -83,12 +79,7 @@ def set_video_codec(
             log.error("Input file does not have a video stream to copy codec from.")
         codec = src.videos[0].codec
 
-    if ctr.vstrict:
-        assert ctr.vcodecs is not None
-        if codec not in ctr.vcodecs:
-            log.error(codec_error.format(codec, out_ext))
-
-    if codec in ctr.disallow_v:
+    if ctr.vcodecs is not None and codec not in ctr.vcodecs:
         log.error(codec_error.format(codec, out_ext))
 
     return codec
@@ -99,8 +90,10 @@ def set_audio_codec(
 ) -> str:
     if codec == "auto":
         codec = "aac" if (src is None or not src.audios) else src.audios[0].codec
-        if ctr.acodecs is not None and codec not in ctr.acodecs:
-            return ctr.acodecs[0]
+        if codec not in ctr.acodecs and ctr.default_aud != "none":
+            return ctr.default_aud
+        if codec == "mp3float":
+            return "mp3"
         return codec
 
     if codec == "copy":
@@ -209,10 +202,8 @@ def edit_media(
     else:
         samplerate = args.sample_rate
 
-    ensure = Ensure(ffmpeg, bar, samplerate, temp, log)
-
     if tl is None:
-        tl = make_timeline(sources, ensure, args, samplerate, bar, temp, log)
+        tl = make_timeline(sources, args, samplerate, bar, temp, log)
 
     if export["export"] == "timeline":
         from auto_editor.formats.json import make_json_timeline
@@ -223,7 +214,7 @@ def edit_media(
     if args.preview:
         from auto_editor.preview import preview
 
-        preview(ensure, tl, temp, log)
+        preview(tl, temp, log)
         return
 
     if export["export"] == "json":
@@ -272,13 +263,15 @@ def edit_media(
         sub_output = []
         apply_later = False
 
-        if ctr.allow_subtitle and not args.sn:
+        ensure = Ensure(ffmpeg, bar, samplerate, temp, log)
+
+        if ctr.default_sub != "none" and not args.sn:
             sub_output = make_new_subtitles(tl, ensure, temp)
 
-        if ctr.allow_audio:
+        if ctr.default_aud != "none":
             audio_output = make_new_audio(tl, ensure, args, ffmpeg, bar, temp, log)
 
-        if ctr.allow_video:
+        if ctr.default_vid != "none":
             if tl.v:
                 out_path, apply_later = render_av(ffmpeg, tl, args, bar, ctr, temp, log)
                 visual_output.append((True, out_path))

auto_editor/lang/palet.py

@@ -18,12 +18,7 @@ from typing import TYPE_CHECKING
 import numpy as np
 from numpy import logical_and, logical_not, logical_or, logical_xor
 
-from auto_editor.analyze import (
-    LevelError,
-    mut_remove_large,
-    mut_remove_small,
-    to_threshold,
-)
+from auto_editor.analyze import LevelError, mut_remove_large, mut_remove_small
 from auto_editor.lib.contracts import *
 from auto_editor.lib.data_structs import *
 from auto_editor.lib.err import MyError
@@ -690,6 +685,9 @@ def palet_map(proc: Proc, seq: Any) -> Any:
         return Quoted(tuple(map(proc, seq.val)))
     if isinstance(seq, list | range):
         return list(map(proc, seq))
+    elif isinstance(seq, np.ndarray):
+        vectorized_proc = np.vectorize(proc)
+        return vectorized_proc(seq)
     return proc(seq)
 
 
@@ -1469,6 +1467,26 @@ def edit_all() -> np.ndarray:
     return env["@levels"].all()
 
 
+def audio_levels(stream: int) -> np.ndarray:
+    if "@levels" not in env:
+        raise MyError("Can't use `audio` if there's no input media")
+
+    try:
+        return env["@levels"].audio(stream)
+    except LevelError as e:
+        raise MyError(e)
+
+
+def motion_levels(stream: int, blur: int = 9, width: int = 400) -> np.ndarray:
+    if "@levels" not in env:
+        raise MyError("Can't use `motion` if there's no input media")
+
+    try:
+        return env["@levels"].motion(stream, blur, width)
+    except LevelError as e:
+        raise MyError(e)
+
+
 def edit_audio(
     threshold: float = 0.04,
     stream: object = Sym("all"),
@@ -1491,7 +1509,7 @@ def edit_audio(
 
     try:
         for s in stream_range:
-            audio_list = to_threshold(levels.audio(s), threshold)
+            audio_list = levels.audio(s) >= threshold
             if stream_data is None:
                 stream_data = audio_list
             else:
@@ -1521,7 +1539,7 @@ def edit_motion(
     levels = env["@levels"]
     strict = env["@filesetup"].strict
     try:
-        return to_threshold(levels.motion(stream, blur, width), threshold)
+        return levels.motion(stream, blur, width) >= threshold
     except LevelError as e:
         return raise_(e) if strict else levels.all()
 
@@ -1582,7 +1600,7 @@ def my_eval(env: Env, node: object) -> Any:
                     return ref(oper, my_eval(env, node[1]))
 
             raise MyError(
-                f"Tried to run: {print_str(oper)} with args: {print_str(node[1:])}"
+                f"{print_str(oper)} is not a function. Tried to run with args: {print_str(node[1:])}"
             )
 
         if type(oper) is Syntax:
@@ -1617,10 +1635,12 @@ env.update({
     # edit procedures
     "none": Proc("none", edit_none, (0, 0)),
     "all/e": Proc("all/e", edit_all, (0, 0)),
+    "audio-levels": Proc("audio-levels", audio_levels, (1, 1), is_nat),
     "audio": Proc("audio", edit_audio, (0, 4),
         is_threshold, orc(is_nat, Sym("all")), is_nat,
         {"threshold": 0, "stream": 1, "minclip": 2, "mincut": 2}
     ),
+    "motion-levels": Proc("motion-levels", motion_levels, (1, 3), is_nat, is_nat1, {"blur": 1, "width": 2}),
     "motion": Proc("motion", edit_motion, (0, 4),
         is_threshold, is_nat, is_nat1,
         {"threshold": 0, "stream": 1, "blur": 1, "width": 2}