yta-video-opengl 0.0.13__tar.gz → 0.0.14__tar.gz

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: yta-video-opengl
-Version: 0.0.13
+Version: 0.0.14
 Summary: Youtube Autonomous Video OpenGL Module
 Author: danialcala94
 Author-email: danielalcalavalera@gmail.com

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "yta-video-opengl"
-version = "0.0.13"
+version = "0.0.14"
 description = "Youtube Autonomous Video OpenGL Module"
 authors = [
     {name = "danialcala94",email = "danielalcalavalera@gmail.com"}

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/src/yta_video_opengl/complete/timeline.py

@@ -218,6 +218,7 @@ class Timeline:
         for t in get_ts(start, end, self.fps):
             frame = self.get_frame_at(t)
 
+            print(f'Getting t:{str(float(t))}')
             #print(frame)
 
             # We need to adjust our output elements to be

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/src/yta_video_opengl/complete/track.py

@@ -94,7 +94,7 @@ class _Part:
             # TODO: By now I'm raising exception to check if
             # this happens or not because I think it would
             # be malfunctioning
-            raise Exception(f'Video is returning None frame at t={str(t)}.')
+            raise Exception(f'Video is returning None video frame at t={str(t)}.')
         
         return frame
     

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/src/yta_video_opengl/complete/video_on_track.py

@@ -150,10 +150,7 @@ class VideoOnTrack:
         )
 
         for frame in frames:
-            # TODO: I am generating a tuple in the
-            # src\yta_video_opengl\reader\cache.py
-            # get_frames method... maybe remove it (?)
-            yield frame[0]
+            yield frame
 
         # # TODO: This was a simple return before
         # return (

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/src/yta_video_opengl/reader/__init__.py

@@ -620,17 +620,6 @@ class VideoReader:
         ):
             yield frame
 
-    # TODO: Will we use this (?)
-    def get_frame(
-        self,
-        index: int
-    ) -> 'VideoFrame':
-        """
-        Get the video frame with the given 'index',
-        using the video cache system.
-        """
-        return self.video_cache.get_frame(index)
-    
     def get_frame_from_t(
         self,
         t: Union[int, float, Fraction]
@@ -639,18 +628,8 @@ class VideoReader:
         Get the video frame with the given 't' time
         moment, using the video cache system.
         """
-        return self.video_cache.get_frame_from_t(t)
+        return self.video_cache.get_video_frame(t)
     
-    def get_audio_frame(
-        self,
-        index: int
-    ) -> 'AudioFrame':
-        """
-        Get the audio frame with the given 'index',
-        using the audio cache system.
-        """
-        return self.audio_cache.get_frame(index)
-
     def get_audio_frame_from_t(
         self,
         t: Union[int, float, Fraction]
@@ -659,7 +638,7 @@ class VideoReader:
         Get the audio frame with the given 't' time
         moment, using the audio cache system.
         """
-        return self.audio_cache.get_frame_from_t(t)
+        return self.audio_cache.get_audio_frame_from_t(t)
     
     def get_audio_frames_from_t(
         self,
@@ -677,7 +656,10 @@ class VideoReader:
         with more than 1 audio frame).
         """
         t: T = T.from_fps(t, self.fps)
-        for frame in self.audio_cache.get_frames(t.truncated, t.next(1).truncated):
+        # We want all the audios that must be played
+        # during the video frame that starts in the
+        # 't' time moment
+        for frame in self.audio_cache.get_audio_frames(t.truncated, t.next(1).truncated):
             yield frame
     
     def get_frames(
@@ -701,7 +683,7 @@ class VideoReader:
         Iterator to get the audio frames in between
         the provided 'start' and 'end' time moments.
         """
-        for frame in self.audio_cache.get_frames(start, end):
+        for frame in self.audio_cache.get_audio_frames(start, end):
             yield frame
     
     def close(

yta_video_opengl-0.0.14/src/yta_video_opengl/reader/cache.py

@@ -0,0 +1,512 @@
+"""
+The pyav container stores the information based
+on the packets timestamps (called 'pts'). Some
+of the packets are considered key_frames because
+they include those key frames.
+
+Also, this library uses those key frames to start
+decodifying from there to the next one, obtaining
+all the frames in between able to be read and
+modified.
+
+This cache system will look for the range of 
+frames that belong to the key frame related to the
+frame we are requesting in the moment, keeping in
+memory all those frames to be handled fast. It
+will remove the old frames if needed to use only
+the 'size' we set when creating it.
+
+A stream can have 'fps = 60' but use another
+different time base that make the pts values go 0,
+ 256, 512... for example. The 'time_base' is the
+only accurate way to obtain the pts.
+
+Feel free to move this explanation to other
+place, its about the duration.
+
+The stream 'duration' parameter is measured
+on ticks, the amount of ticks that the
+stream lasts. Here below is an example:
+
+- Duration raw: 529200
+- Time base: 1/44100
+- Duration (seconds): 12.0
+"""
+from yta_video_opengl.t import T
+from av.container import InputContainer
+from av.video.stream import VideoStream
+from av.audio.stream import AudioStream
+from av.video.frame import VideoFrame
+from av.audio.frame import AudioFrame
+from av.packet import Packet
+from yta_validation.parameter import ParameterValidator
+from yta_validation import PythonValidator
+from quicktions import Fraction
+from collections import OrderedDict
+from typing import Union
+
+import numpy as np
+import math
+
+
+# TODO: This is not actually a Video
+# cache, is a FrameCache because we
+# create one for video but another
+# one for audio. Rename it please.
+class VideoFrameCache:
+    """
+    Class to manage the frames cache of a video
+    within a video reader instance.
+    """
+
+    @property
+    def fps(
+        self
+    ) -> Union[int, Fraction, None]:
+        """
+        The frames per second.
+        """
+        return (
+            self.stream.average_rate
+            if self.stream.type == 'video' else
+            self.stream.rate
+        )
+    
+    @property
+    def time_base(
+        self
+    ) -> Union[Fraction, None]:
+        """
+        The time base of the stream.
+        """
+        return self.stream.time_base
+
+    def __init__(
+        self,
+        container: InputContainer,
+        stream: Union[VideoStream, AudioStream],
+        size: Union[int, None] = None
+    ):
+        ParameterValidator.validate_mandatory_instance_of('container', container, InputContainer)
+        ParameterValidator.validate_mandatory_instance_of('stream', stream, [VideoStream, AudioStream])
+        ParameterValidator.validate_positive_int('size', size)
+
+        self.container: InputContainer = container
+        """
+        The pyav container.
+        """
+        self.stream: Union[VideoStream, AudioStream] = stream
+        """
+        The pyav stream.
+        """
+        self.cache: OrderedDict = OrderedDict()
+        """
+        The cache ordered dictionary.
+        """
+        self.size: Union[int, None] = size
+        """
+        The size (in number of frames) of the cache.
+        """
+        self.key_frames_pts: list[int] = []
+        """
+        The list that contains the timestamps of the
+        key frame packets, ordered from begining to
+        end.
+        """
+
+        # TODO: This is new, remove this comment if
+        # it is ok
+        # TODO: This way of obtaining the duration
+        # in ticks must be a utils
+        self.frame_duration: int = (
+            self.stream.duration / self.stream.frames
+            if PythonValidator.is_instance_of(stream, VideoStream) else
+            # TODO: Is this below ok (?)
+            self.stream.frames
+        )
+        """
+        The duration (in ticks) of the frame, that
+        is the step between the different pts.
+        """
+        self._last_packet_accessed: Union[Packet, None] = None
+        """
+        The last packet that has been accessed
+        """
+        self._last_frame_read: Union[VideoFrame, AudioFrame, None] = None
+        """
+        The last frame we have read when decoding.
+        Useful to avoid seeking all the time when we
+        don't need it.
+        """
+
+        self._prepare()
+
+    def _prepare(
+        self
+    ):
+        # Index key frames
+        for packet in self.container.demux(self.stream):
+            if packet.is_keyframe:
+                self.key_frames_pts.append(packet.pts)
+
+        # The cache size will be auto-calculated to
+        # use the amount of frames of the biggest
+        # interval of frames that belongs to a key
+        # frame, or a value by default
+        # TODO: Careful if this is too big
+        fps = (
+            float(self.stream.average_rate)
+            if PythonValidator.is_instance_of(self.stream, VideoStream) else
+            float(self.stream.rate)
+        )
+        # Intervals, but in number of frames
+        intervals = np.diff(
+            # Intervals of time between keyframes
+            np.array(self.key_frames_pts) * self.time_base
+        ) * fps
+
+        self.size = (
+            math.ceil(np.max(intervals))
+            if intervals.size > 0 else
+            (
+                self.size or
+                # TODO: Make this 'default_size' a setting or something
+                60
+            )
+        )
+        
+        self.container.seek(0)
+
+    def _get_nearest_keyframe_pts(
+        self,
+        pts: int
+    ):
+        """
+        Get the fps of the keyframe that is the
+        nearest to the provided 'pts'. Useful to
+        seek and start decoding frames from that
+        keyframe.
+        """
+        return max([
+            key_frame_pts
+            for key_frame_pts in self.key_frames_pts
+            if key_frame_pts <= pts
+        ])
+
+    def _store_frame_in_cache(
+        self,
+        frame: Union[VideoFrame, AudioFrame]
+    ) -> Union[VideoFrame, AudioFrame]:
+        """
+        Store the provided 'frame' in cache if it
+        is not on it, removing the first item of
+        the cache if full.
+        """
+        if frame.pts not in self.cache:
+            self.cache[frame.pts] = frame
+
+            # Clean cache if full
+            if len(self.cache) > self.size:
+                self.cache.popitem(last = False)
+
+        return frame
+    
+    def _seek(
+        self,
+        pts: int
+    ):
+        """
+        Seek to the given 'pts' only if it is not
+        the next 'pts' to the last read, and it 
+        will also apply a pad to avoid problems
+        when reading audio frames.
+
+        TODO: Apply the padding only to audio 
+        frame reading (?)
+        """
+        # I found that it is recommended to
+        # read ~100ms before the pts we want to
+        # actually read so we obtain the frames
+        # clean (this is important in audio)
+        # TODO: This is maybe too much for a
+        # video and not needed
+        pts_pad = int(0.1 / self.time_base)
+        self.container.seek(
+            offset = max(0, pts - pts_pad),
+            stream = self.stream
+        )
+
+    def get_video_frame(
+        self,
+        t: Union[int, float, Fraction]
+    ) -> VideoFrame:
+        """
+        Get the video frame that is in the 't'
+        time moment provided.
+        """
+        for frame in self.get_video_frames(t):
+            return frame
+
+    def get_video_frames(
+        self,
+        start: Union[int, float, Fraction] = 0,
+        end: Union[int, float, Fraction, None] = None
+    ):
+        """
+        Get all the frames in the range between
+        the provided 'start' and 'end' time in
+        seconds.
+
+        This method is an iterator that yields
+        the frame, its t and its index.
+        """
+        start = T(start, self.time_base).truncated
+        end = (
+            T(end, self.time_base).truncated
+            if end is not None else
+            # The next frame
+            start + (1 / self.fps)
+        )
+
+        key_frame_pts = self._get_nearest_keyframe_pts(start / self.time_base)
+
+        if (
+            self._last_packet_accessed is None or
+            self._last_packet_accessed.pts != key_frame_pts
+        ):
+            self._seek(key_frame_pts)
+
+        for packet in self.container.demux(self.stream):
+            if packet.pts is None:
+                continue
+
+            self._last_packet_accessed = packet
+
+            for frame in packet.decode():
+                if frame.pts is None:
+                    continue
+
+                # We store all the frames in cache
+                self._store_frame_in_cache(frame)
+                
+                current_frame_time = frame.pts * self.time_base
+                
+                # We want the range [start, end)
+                if start <= current_frame_time < end:
+                    yield frame
+
+                if current_frame_time >= end:
+                    break
+
+    def get_audio_frame_from_t(
+        self,
+        t: Union[int, float, Fraction]
+    ):
+        """
+        Get the single audio frame that must be
+        played at the 't' time moment provided.
+        This method is useful to get the single
+        audio frame that we need to combine 
+        when using it in a composition.
+
+        TODO: Are we actually using this method (?)
+        """
+        t: T = T(t, self.time_base)
+        # We need the just one audio frame
+        for frame in self.get_audio_frames(t.truncated, t.next(1).truncated):
+            return frame
+
+    def get_audio_frames_from_t(
+        self,
+        t: Union[int, float, Fraction]
+    ):
+        """
+        Get all the audio frames that must be
+        played at the 't' time moment provided.
+        """
+        for frame in self.get_audio_frames(t):
+            yield frame
+
+    def get_audio_frames(
+        self,
+        start: Union[int, float, Fraction] = 0,
+        end: Union[int, float, Fraction, None] = None
+    ):
+        """
+        Get all the audio frames in the range
+        between the provided 'start' and 'end'
+        time (in seconds).
+
+        This method is an iterator that yields
+        the frame, its t and its index.
+        """
+        # TODO: Is this ok? We are trying to obtain
+        # the audio frames for a video frame, so
+        # should we use the 'self.time_base' to
+        # truncate (?)
+        start = T(start, self.time_base).truncated
+        end = (
+            T(end, self.time_base).truncated
+            if end is not None else
+            start + (1 / self.fps)
+        )
+
+        key_frame_pts = self._get_nearest_keyframe_pts(start / self.time_base)
+
+        if (
+            self._last_packet_accessed is None or
+            self._last_packet_accessed.pts != key_frame_pts
+        ):
+            self._seek(key_frame_pts)
+
+        for packet in self.container.demux(self.stream):
+            if packet.pts is None:
+                continue
+
+            self._last_packet_accessed = packet
+
+            for frame in packet.decode():
+                if frame.pts is None:
+                    continue
+
+                # We store all the frames in cache
+                self._store_frame_in_cache(frame)
+
+                current_frame_time = frame.pts * self.time_base
+                # End is not included, its the start of the
+                # next frame actually
+                frame_end = current_frame_time + (frame.samples / self.stream.sample_rate)
+
+                # For the next comments imagine we are looking
+                # for the [1.0, 2.0) audio time range
+                # Previous frame and nothing is inside
+                if frame_end <= start:
+                    # From 0.25 to 1.0
+                    continue
+                
+                # We finished, nothing is inside and its after
+                if current_frame_time >= end:
+                    # From 2.0 to 2.75
+                    return
+
+                # If we need audio from 1 to 2, audio is:
+                #   - from 0 to 0.75    (Not included, omit)
+                #   - from 0.5 to 1.5   (Included, take 1.0 to 1.5)
+                #   - from 0.5 to 2.5   (Included, take 1.0 to 2.0)
+                #   - from 1.25 to 1.5  (Included, take 1.25 to 1.5)
+                #   - from 1.25 to 2.5  (Included, take 1.25 to 2.0)
+                #   - from 2.5 to 3.5   (Not included, omit)
+                
+                # Here below, at least a part is inside
+                if (
+                    current_frame_time < start and
+                    frame_end > start
+                ):
+                    # A part at the end is included
+                    end_time = (
+                        # From 0.5 to 1.5 0> take 1.0 to 1.5
+                        frame_end
+                        if frame_end <= end else
+                        # From 0.5 to 2.5 => take 1.0 to 2.0
+                        end
+                    )
+                    #print('A part at the end is included.')
+                    frame = trim_audio_frame(
+                        frame = frame,
+                        start = start,
+                        end = end_time,
+                        time_base = self.time_base
+                    )
+                elif (
+                    current_frame_time >= start and
+                    current_frame_time < end
+                ):
+                    end_time = (
+                        # From 1.25 to 1.5 => take 1.25 to 1.5
+                        frame_end
+                        if frame_end <= end else
+                        # From 1.25 to 2.5 => take 1.25 to 2.0
+                        end
+                    )
+                    # A part at the begining is included
+                    #print('A part at the begining is included.')
+                    frame = trim_audio_frame(
+                        frame = frame,
+                        start = current_frame_time,
+                        end = end_time,
+                        time_base = self.time_base
+                    )
+
+                # If the whole frame is in, past as it is
+                yield frame
+    
+    def clear(
+        self
+    ) -> 'VideoFrameCache':
+        """
+        Clear the cache by removing all the items.
+        """
+        self.cache.clear()
+
+        return self
+
+def trim_audio_frame(
+    frame: AudioFrame,
+    start: Union[int, float, Fraction],
+    end: Union[int, float, Fraction],
+    time_base: Fraction
+) -> AudioFrame:
+    """
+    Trim an audio frame to obtain the part between
+    [start, end), that is provided in seconds.
+    """
+    # (channels, n_samples)
+    samples = frame.to_ndarray()  
+    n_samples = samples.shape[1]
+
+    # In seconds
+    frame_start = frame.pts * float(time_base)
+    frame_end = frame_start + (n_samples / frame.sample_rate)
+
+    # Overlapping 
+    cut_start = max(frame_start, float(start))
+    cut_end = min(frame_end, float(end))
+
+    if cut_start >= cut_end:
+        # No overlapping
+        return None  
+
+    # To sample indexes
+    start_index = int(round((cut_start - frame_start) * frame.sample_rate))
+    end_index = int(round((cut_end - frame_start) * frame.sample_rate))
+
+    new_frame = AudioFrame.from_ndarray(
+        # end_index is not included: so [start, end)
+        array = samples[:, start_index:end_index],
+        format = frame.format,
+        layout = frame.layout
+    )
+
+    # Set attributes
+    new_frame.sample_rate = frame.sample_rate
+    new_frame.time_base = time_base
+    new_frame.pts = int(round(cut_start / float(time_base)))
+
+    return new_frame
+
+
+
+"""
+There is a way of editing videos being
+able to arbitrary access to frames, that
+is transforming the source videos to
+intra-frame videos. This is a ffmpeg
+command that can do it:
+
+- `ffmpeg -i input.mp4 -c:v libx264 -x264opts keyint=1 -preset fast -crf 18 -c:a copy output_intra.mp4`
+
+Once you have the 'output_intra.mp4',
+each packet can decodify its frame 
+depending not on the previous one, being
+able to seek and jump easy.
+"""

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/src/yta_video_opengl/t.py

@@ -1,3 +1,14 @@
+"""
+This is an example of what a video has:
+- fps = 60
+- time_base = 1 / 15360
+- tick = fps * time_base = 256
+
+So, the first pts is 0 and the second 
+one is 256. The frame 16 will be 3840,
+that is 256 * 15 (because first index
+is 0).
+"""
 from yta_validation.parameter import ParameterValidator
 from yta_validation import PythonValidator
 from yta_validation.number import NumberValidator
@@ -36,7 +47,7 @@ class T:
         The 't' but as a Fraction that is multiple
         of the given 'time_base' and rounded (the
         value could be the same as truncated if it
-        is closer to the previou value).
+        is closer to the previous value).
         """
         return round_t(self._t, self.time_base, do_truncate = False)
     
@@ -96,20 +107,57 @@ class T:
         """
         return T(self.truncated + n * self.time_base, self.time_base)
     
-    # TODO: Maybe its better to make the '__init__'
-    # receive the fps and create the 'from_time_base'
-    # because I think we will provide the fps or the
-    # sample rate more often
+    def previous(
+        self,
+        n: int = 1
+    ) -> 'T':
+        """
+        Get the value that is 'n' times before the
+        'truncated' property of this instance.
+
+        Useful when you need the previous value to
+        check if the current is the next one or
+        similar.
+
+        Be careful, if the 'truncated' value is 0
+        this will give you an unexpected negative
+        value.
+        """
+        return T(self.truncated - n * self.time_base, self.time_base)
+    
     @staticmethod
     def from_fps(
         t: Union[int, float, Fraction],
         fps: Union[int, float, Fraction]
-    ):
+    ) -> 'T':
         """
         Get the instance but providing the 'fps'
-        (or sample rate) value directly.
+        (or sample rate) value directly, that will
+        be turned into a time base.
         """
         return T(t, fps_to_time_base(fps))
+
+    @staticmethod
+    def from_pts(
+        pts: int,
+        time_base: Fraction
+    ) -> 'T':
+        """
+        Get the instance but providing the 'pts'
+        and the 'time_base'.
+        """
+        return T(pts * time_base, time_base)
+    
+
+# TODO: Careful with this below
+"""
+To obtain the pts step, or frame duration in
+ticks, you need to apply 2 formulas that are
+different according to if the frame is video
+or audio:
+- Audio: .samples
+- Video: int(round((1 / .fps) / .time_base))
+"""
     
 def get_ts(
     start: Union[int, float, Fraction],

{yta_video_opengl-0.0.13 → yta_video_opengl-0.0.14}/src/yta_video_opengl/video.py

@@ -182,15 +182,20 @@ class Video:
         Get the video frame with the given 't' time
         moment, using the video cache system.
         """
-        return self.reader.video_cache.get_frame_from_t(self._get_real_t(t))
-    
+        return self.reader.video_cache.get_video_frame(self._get_real_t(t))
+
     def get_audio_frame_from_t(
         self,
         t: Union[int, float, Fraction]
     ) -> 'AudioFrame':
         """
         Get the audio frame with the given 't' time
-        moment, using the audio cache system.
+        moment, using the audio cache system. This
+        method is useful when we need to combine 
+        many different frames so we can obtain them
+        one by one.
+
+        TODO: Is this actually necessary (?)
         """
         return self.reader.audio_cache.get_frame_from_t(self._get_real_t(t))
     

yta_video_opengl-0.0.13/src/yta_video_opengl/reader/cache.py

@@ -1,529 +0,0 @@
-"""
-The pyav container stores the information based
-on the packets timestamps (called 'pts'). Some
-of the packets are considered key_frames because
-they include those key frames.
-
-Also, this library uses those key frames to start
-decodifying from there to the next one, obtaining
-all the frames in between able to be read and
-modified.
-
-This cache system will look for the range of 
-frames that belong to the key frame related to the
-frame we are requesting in the moment, keeping in
-memory all those frames to be handled fast. It
-will remove the old frames if needed to use only
-the 'size' we set when creating it.
-
-A stream can have 'fps = 60' but use another
-different time base that make the pts values go 0,
- 256, 512... for example. The 'time_base' is the
-only accurate way to obtain the pts.
-"""
-from yta_video_opengl.utils import t_to_pts, pts_to_t, pts_to_index, index_to_pts
-from yta_video_opengl.t import T
-from av.container import InputContainer
-from av.video.stream import VideoStream
-from av.audio.stream import AudioStream
-from av.video.frame import VideoFrame
-from av.audio.frame import AudioFrame
-from yta_validation.parameter import ParameterValidator
-from yta_validation import PythonValidator
-from quicktions import Fraction
-from collections import OrderedDict
-from typing import Union
-
-import numpy as np
-import av
-import math
-
-
-# TODO: This is not actually a Video
-# cache, is a FrameCache because we
-# create one for video but another
-# one for audio. Rename it please.
-class VideoFrameCache:
-    """
-    Class to manage the frames cache of a video
-    within a video reader instance.
-    """
-
-    @property
-    def fps(
-        self
-    ) -> float:
-        """
-        The frames per second as a float.
-        """
-        return (
-            float(self.stream.average_rate)
-            if self.stream.type == 'video' else
-            float(self.stream.rate)
-        )
-    
-    @property
-    def time_base(
-        self
-    ) -> Union[Fraction, None]:
-        """
-        The time base of the stream.
-        """
-        return self.stream.time_base
-
-    def __init__(
-        self,
-        container: InputContainer,
-        stream: Union[VideoStream, AudioStream],
-        size: Union[int, None] = None
-    ):
-        ParameterValidator.validate_mandatory_instance_of('container', container, InputContainer)
-        ParameterValidator.validate_mandatory_instance_of('stream', stream, [VideoStream, AudioStream])
-        ParameterValidator.validate_positive_int('size', size)
-
-        self.container: InputContainer = container
-        """
-        The pyav container.
-        """
-        self.stream: Union[VideoStream, AudioStream] = stream
-        """
-        The pyav stream.
-        """
-        self.cache: OrderedDict = OrderedDict()
-        """
-        The cache ordered dictionary.
-        """
-        self.size: Union[int, None] = size
-        """
-        The size (in number of frames) of the cache.
-        """
-        self.key_frames_pts: list[int] = []
-        """
-        The list that contains the timestamps of the
-        key frame packets, ordered from begining to
-        end.
-        """
-
-        self._prepare()
-
-    def _prepare(
-        self
-    ):
-        # Index key frames
-        for packet in self.container.demux(self.stream):
-            if packet.is_keyframe:
-                self.key_frames_pts.append(packet.pts)
-
-        # The cache size will be auto-calculated to
-        # use the amount of frames of the biggest
-        # interval of frames that belongs to a key
-        # frame, or a value by default
-        # TODO: Careful if this is too big
-        fps = (
-            float(self.stream.average_rate)
-            if PythonValidator.is_instance_of(self.stream, VideoStream) else
-            float(self.stream.rate)
-        )
-        # Intervals, but in number of frames
-        intervals = np.diff(
-            # Intervals of time between keyframes
-            np.array(self.key_frames_pts) * self.time_base
-        ) * fps
-
-        self.size = (
-            math.ceil(np.max(intervals))
-            if intervals.size > 0 else
-            (
-                self.size or
-                # TODO: Make this 'default_size' a setting or something
-                60
-            )
-        )
-        
-        self.container.seek(0)
-
-    def _get_nearest_keyframe_pts(
-        self,
-        pts: int
-    ):
-        """
-        Get the fps of the keyframe that is the
-        nearest to the provided 'pts'. Useful to
-        seek and start decoding frames from that
-        keyframe.
-        """
-        return max([
-            key_frame_pts
-            for key_frame_pts in self.key_frames_pts
-            if key_frame_pts <= pts
-        ])
-
-    def _store_frame_in_cache(
-        self,
-        frame: Union[VideoFrame, AudioFrame]
-    ) -> Union[VideoFrame, AudioFrame]:
-        """
-        Store the provided 'frame' in cache if it
-        is not on it, removing the first item of
-        the cache if full.
-        """
-        if frame.pts not in self.cache:
-            self.cache[frame.pts] = frame
-
-            # Clean cache if full
-            if len(self.cache) > self.size:
-                self.cache.popitem(last = False)
-
-        return frame
-
-    def get_frame_from_pts(
-        self,
-        pts: int
-    ) -> Union[VideoFrame, AudioFrame, None]:
-        """
-        Get the frame that has the provided 'pts'.
-
-        This method will start decoding frames from the
-        most near key frame (the one with the nearer
-        pts) until the one requested is found. All those
-        frames will be stored in cache.
-
-        This method must be called when the frame 
-        requested is not stored in the caché.
-        """
-        if pts in self.cache:
-            return self.cache[pts]
-        
-        # Look for the most near key frame
-        key_frame_pts = self._get_nearest_keyframe_pts(pts)
-
-        # Go to the key frame that includes it
-        # but I read that it is recommended to
-        # read ~100ms before the pts we want to
-        # actually read so we obtain the frames
-        # clean (this is important in audio)
-        # TODO: This code is repeated, refactor
-        pts_pad = int(0.1 / self.time_base)
-        self.container.seek(
-            offset = max(0, key_frame_pts - pts_pad),
-            stream = self.stream
-        )
-
-        decoded = None
-        for frame in self.container.decode(self.stream):
-            # TODO: Could 'frame' be None (?)
-            if frame.pts is None:
-                continue
-
-            # Store in cache if needed
-            self._store_frame_in_cache(frame)
-
-            """
-            The 'frame.pts * frame.time_base' will give
-            us the index of the frame, and actually the
-            'pts' que are looking for seems to be the
-            index and not a pts.
-
-            TODO: Review all this in all the logic 
-            please.
-            """
-            if frame.pts >= pts:
-                decoded = self.cache[frame.pts]
-                break
-
-        # TODO: Is this working? We need previous 
-        # frames to be able to decode...
-        return decoded
-
-    # TODO: I'm not using this method...
-    def get_frame(
-        self,
-        index: int
-    ) -> Union[VideoFrame, AudioFrame]:
-        """
-        Get the frame with the given 'index' from
-        the cache.
-        """
-        # TODO: Maybe we can accept 'pts' also
-        pts = index_to_pts(index, self.time_base, self.fps)
-
-        return (
-            self.cache[pts]
-            if pts in self.cache else
-            self.get_frame_from_pts(pts)
-        )
-    
-    def get_frame_from_t(
-        self,
-        t: Union[int, float, Fraction]
-    ) -> Union[VideoFrame, AudioFrame]:
-        """
-        Get the frame with the given 't' time moment
-        from the cache.
-        """
-        return self.get_frame_from_pts(T(t, self.time_base).truncated_pts)
-
-    def get_frames(
-        self,
-        start: Union[int, float, Fraction] = 0,
-        end: Union[int, float, Fraction, None] = None
-    ):
-        """
-        Get all the frames in the range between
-        the provided 'start' and 'end' time in
-        seconds.
-
-        This method is an iterator that yields
-        the frame, its t and its index.
-        """
-        # We use the cache as iterator if all the frames
-        # requested are stored there
-        # TODO: I think this is not ok... I will never
-        # have all the pts form here stored, as they come
-        # from 't' that is different...
-
-        """
-        Feel free to move this explanation to other
-        place, its about the duration.
-
-        The stream 'duration' parameter is measured
-        on ticks, the amount of ticks that the
-        stream lasts. Here below is an example:
-
-        - Duration raw: 529200
-        - Time base: 1/44100
-        - Duration (seconds): 12.0
-        """
-
-        # The 'duration' is on pts ticks
-        duration = float(self.stream.duration * self.time_base)
-        # TODO: I think it would be better to
-        # receive and work with pts instead of
-        # 't' time moments...
-        # pts_list = [
-        #     t_to_pts(t, self.time_base)
-        #     for t in T.get_frame_indexes(duration, self.fps, start, end)
-        # ]
-
-        # if all(
-        #     pts in self.cache
-        #     for pts in pts_list
-        # ):
-        #     for pts in pts_list:
-        #         yield self.cache[pts]
-
-        # If not all, we ignore the cache because we
-        # need to decode and they are all consecutive
-        start = T(start, self.time_base).truncated_pts
-        end = (
-            T(end, self.time_base).truncated_pts
-            if end is not None else
-            None
-        )
-        key_frame_pts = self._get_nearest_keyframe_pts(start)
-
-        # Go to the key frame that includes it
-        # but I read that it is recommended to
-        # read ~100ms before the pts we want to
-        # actually read so we obtain the frames
-        # clean (this is important in audio)
-        # TODO: This code is repeated, refactor
-        pts_pad = int(0.1 / self.time_base)
-        self.container.seek(
-            offset = max(0, key_frame_pts - pts_pad),
-            stream = self.stream
-        )
-
-        for packet in self.container.demux(self.stream):
-            for frame in packet.decode():
-                if frame.pts is None:
-                    continue
-
-                # We store all the frames in cache
-                self._store_frame_in_cache(frame)
-
-                frame_end_pts = frame.pts + int(frame.samples * (1 / self.stream.sample_rate) / self.time_base)
-                #frame_end_pts = frame.pts + int(frame.samples)
-                #frame_end_pts = frame.pts + int(frame.samples / (self.stream.sample_rate * self.time_base))
-
-                # For the next comments imagine we are looking
-                # for the [1.0, 2.0) audio time range
-                # Previous frame and nothing is inside
-                if frame_end_pts <= start:
-                    # From 0.25 to 1.0
-                    continue
-
-                # We finished, nothing is inside and its after
-                if (
-                    end is not None and
-                    frame.pts >= end
-                ):
-                    # From 2.0 to 2.75
-                    return
-
-                # We need: from 1 to 2
-                # Audio is:
-                #   - from 0 to 0.75    (Not included, omit)
-                #   - from 0.5 to 1.5   (Included, take 1.0 to 1.5)
-                #   - from 0.5 to 2.5   (Included, take 1.0 to 2.0)
-                #   - from 1.25 to 1.5  (Included, take 1.25 to 1.5)
-                #   - from 1.25 to 2.5  (Included, take 1.25 to 2.0)
-                #   - from 2.5 to 3.5   (Not included, omit)
-                
-                # Here below, at least a part is inside
-                if (
-                    frame.pts < start and
-                    frame_end_pts > start
-                ):
-                    # A part at the end is included
-                    end_time = (
-                        # From 0.5 to 1.5 0> take 1.0 to 1.5
-                        frame_end_pts
-                        if frame_end_pts <= end else
-                        # From 0.5 to 2.5 => take 1.0 to 2.0
-                        end
-                    )
-                    #print('A part at the end is included.')
-                    # TODO: I'm using too much 'pts_to_t'
-                    frame = trim_audio_frame_pts(
-                        frame = frame,
-                        start_pts = start,
-                        end_pts = end_time,
-                        time_base = self.time_base
-                    )
-                elif (
-                    frame.pts >= start and
-                    frame.pts < end
-                ):
-                    end_time = (
-                        # From 1.25 to 1.5 => take 1.25 to 1.5
-                        frame_end_pts
-                        if frame_end_pts <= end else
-                        # From 1.25 to 2.5 => take 1.25 to 2.0
-                        end
-                    )
-                    # A part at the begining is included
-                    #print('A part at the begining is included.')
-                    # TODO: I'm using too much 'pts_to_t'
-                    frame = trim_audio_frame_pts(
-                        frame = frame,
-                        start_pts = frame.pts,
-                        end_pts = end_time,
-                        time_base = self.time_base
-                    )
-
-                # If the whole frame is in, past as it is
-                
-                # TODO: Maybe send a @dataclass instead (?)
-                # TODO: Do I really need these 't' and 'index' (?)
-                yield (
-                    frame,
-                    pts_to_t(frame.pts, self.time_base),
-                    pts_to_index(frame.pts, self.time_base, self.fps)
-                )
-    
-    def clear(
-        self
-    ) -> 'VideoFrameCache':
-        """
-        Clear the cache by removing all the items.
-        """
-        self.cache.clear()
-
-        return self
-    
-
-# TODO: Move this to a utils when refactored
-def trim_audio_frame_pts(
-    frame: av.AudioFrame,
-    start_pts: int,
-    end_pts: int,
-    time_base
-) -> av.AudioFrame:
-    """
-    Recorta un AudioFrame para quedarse solo con la parte entre [start_pts, end_pts] en ticks (PTS).
-    """
-    samples = frame.to_ndarray()  # (channels, n_samples)
-    n_channels, n_samples = samples.shape
-    sr = frame.sample_rate
-
-    #frame_end_pts = frame.pts + int((n_samples / sr) / time_base)
-    # TODO: This could be wrong
-    frame_end_pts = frame.pts + int(frame.samples)
-
-    # solapamiento en PTS
-    cut_start_pts = max(frame.pts, start_pts)
-    cut_end_pts = min(frame_end_pts, end_pts)
-
-    if cut_start_pts >= cut_end_pts:
-        raise Exception('Oops...')
-        return None  # no hay solapamiento
-
-    # convertir a índices de samples (en ticks → segundos → samples)
-    cut_start_time = (cut_start_pts - frame.pts) * time_base
-    cut_end_time = (cut_end_pts - frame.pts) * time_base
-
-    start_idx = int(cut_start_time * sr)
-    end_idx = int(cut_end_time * sr)
-
-    # print(
-    #     f"cutting [{frame.pts}, {frame_end_pts}] "
-    #     f"to [{cut_start_pts}, {cut_end_pts}] "
-    #     f"({start_idx}:{end_idx} / {frame.samples})"
-    #     #f"({start_idx}:{end_idx} / {n_samples})"
-    # )
-
-    cut_samples = samples[:, start_idx:end_idx]
-
-    # crear nuevo AudioFrame
-    new_frame = av.AudioFrame.from_ndarray(cut_samples, format=frame.format, layout=frame.layout)
-    new_frame.sample_rate = sr
-
-    # ajustar PTS → corresponde al inicio real del recorte
-    new_frame.pts = cut_start_pts
-    new_frame.time_base = time_base
-
-    return new_frame
-
-
-
-def trim_audio_frame_t(
-    frame: av.AudioFrame,
-    start_time: float,
-    end_time: float,
-    time_base
-) -> av.AudioFrame:
-    """
-    Recorta un AudioFrame para quedarse solo con la parte entre [start_time, end_time] en segundos.
-    """
-    samples = frame.to_ndarray()  # (channels, n_samples)
-    n_channels, n_samples = samples.shape
-    sr = frame.sample_rate
-
-    frame_start = float(frame.pts * time_base)
-    frame_end = frame_start + (n_samples / sr)
-
-    # calcular solapamiento en segundos
-    cut_start = max(frame_start, start_time)
-    cut_end = min(frame_end, end_time)
-
-    if cut_start >= cut_end:
-        return None  # no hay solapamiento
-
-    # convertir a índices de samples
-    start_idx = int((cut_start - frame_start) * sr)
-    end_idx = int((cut_end - frame_start) * sr)
-
-    # print(f'cutting [{str(frame_start)}, {str(frame_end)}] to [{str(float(start_time))}, {str(float(end_time))}] from {str(start_idx)} to {str(end_idx)} of {str(int((frame_end - frame_start) * sr))}')
-    cut_samples = samples[:, start_idx:end_idx]
-
-    # crear nuevo AudioFrame
-    new_frame = av.AudioFrame.from_ndarray(cut_samples, format = frame.format, layout = frame.layout)
-    new_frame.sample_rate = sr
-
-    # ajustar PTS → corresponde al inicio real del recorte
-    new_pts = int(cut_start / time_base)
-    new_frame.pts = new_pts
-    new_frame.time_base = time_base
-
-    return new_frame