videopython 0.26.1__tar.gz → 0.26.3__tar.gz

{videopython-0.26.1 → videopython-0.26.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: videopython
-Version: 0.26.1
+Version: 0.26.3
 Summary: Minimal video generation and processing library.
 Project-URL: Homepage, https://videopython.com
 Project-URL: Repository, https://github.com/bartwojtowicz/videopython/

{videopython-0.26.1 → videopython-0.26.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "videopython"
-version = "0.26.1"
+version = "0.26.3"
 description = "Minimal video generation and processing library."
 authors = [
     { name = "Bartosz Wójtowicz", email = "bartoszwojtowicz@outlook.com" },

{videopython-0.26.1 → videopython-0.26.3}/src/videopython/ai/_device.py

@@ -25,6 +25,33 @@ def log_device_initialization(
     )
 
 
+def release_device_memory(device: str | None) -> None:
+    """Release cached allocator memory for the given device.
+
+    Safe to call when torch is not importable or the device is CPU/None.
+    """
+    try:
+        import torch
+    except ImportError:
+        return
+
+    import gc
+
+    gc.collect()
+
+    if device == "cuda" and torch.cuda.is_available():
+        torch.cuda.empty_cache()
+        return
+
+    if device == "mps":
+        mps_backend = getattr(torch.backends, "mps", None)
+        if mps_backend is not None and mps_backend.is_available():
+            mps_mod = getattr(torch, "mps", None)
+            empty_cache = getattr(mps_mod, "empty_cache", None) if mps_mod is not None else None
+            if callable(empty_cache):
+                empty_cache()
+
+
 def select_device(
     device: str | None,
     *,

{videopython-0.26.1 → videopython-0.26.3}/src/videopython/ai/dubbing/dubber.py

@@ -3,6 +3,8 @@
 from __future__ import annotations
 
 import logging
+import tempfile
+from pathlib import Path
 from typing import TYPE_CHECKING, Any, Callable
 
 from videopython.ai.dubbing.models import DubbingResult, RevoiceResult
@@ -14,18 +16,28 @@ logger = logging.getLogger(__name__)
 
 
 class VideoDubber:
-    """Dubs videos into different languages using the local pipeline."""
-
-    def __init__(self, device: str | None = None):
+    """Dubs videos into different languages using the local pipeline.
+
+    Args:
+        device: Execution device (``cpu``, ``cuda``, ``mps``, or ``None`` for auto).
+        low_memory: When True, each pipeline stage (Whisper, Demucs, MarianMT,
+            Chatterbox TTS) is unloaded from memory after it runs, so only one
+            model is resident at a time. Trades per-run latency (~10-30s of
+            extra model loads) for a much lower memory ceiling. Recommended for
+            GPUs with <=12GB VRAM or hosts with <32GB RAM. Default False.
+    """
+
+    def __init__(self, device: str | None = None, low_memory: bool = False):
         self.device = device
+        self.low_memory = low_memory
         self._local_pipeline: Any = None
         requested = device.lower() if isinstance(device, str) else "auto"
-        logger.info("VideoDubber initialized with device=%s", requested)
+        logger.info("VideoDubber initialized with device=%s low_memory=%s", requested, low_memory)
 
     def _init_local_pipeline(self) -> None:
         from videopython.ai.dubbing.pipeline import LocalDubbingPipeline
 
-        self._local_pipeline = LocalDubbingPipeline(device=self.device)
+        self._local_pipeline = LocalDubbingPipeline(device=self.device, low_memory=self.low_memory)
 
     def dub(
         self,
@@ -50,7 +62,7 @@ class VideoDubber:
             self._init_local_pipeline()
 
         return self._local_pipeline.process(
-            video=video,
+            source_audio=video.audio,
             target_lang=target_lang,
             source_lang=source_lang,
             preserve_background=preserve_background,
@@ -89,6 +101,84 @@ class VideoDubber:
         )
         return video.add_audio(result.dubbed_audio, overlay=False)
 
+    def dub_file(
+        self,
+        input_path: str | Path,
+        output_path: str | Path,
+        target_lang: str,
+        source_lang: str | None = None,
+        preserve_background: bool = True,
+        voice_clone: bool = True,
+        enable_diarization: bool = False,
+        progress_callback: Callable[[str, float], None] | None = None,
+        transcription: Any = None,
+    ) -> DubbingResult:
+        """Dub a video file in place on disk without loading video frames into memory.
+
+        Extracts the audio track via ffmpeg, runs the dubbing pipeline on the
+        audio only, then muxes the dubbed audio back into the source video
+        using ffmpeg stream-copy (no video re-encode). Peak memory is bounded
+        by model weights and the audio track — independent of video length and
+        resolution.
+
+        Use this instead of ``dub_and_replace`` when the source video is long
+        or high-resolution and you don't need frame-level access in Python.
+
+        Args:
+            input_path: Path to the source video file.
+            output_path: Path to write the dubbed video. Overwritten if it exists.
+            target_lang: Target language code (e.g. ``"es"``, ``"fr"``).
+            source_lang: Source language code, or ``None`` to auto-detect.
+            preserve_background: Preserve background music/effects via source separation.
+            voice_clone: Clone the source speaker's voice for the dubbed track.
+            enable_diarization: Enable speaker diarization for per-speaker voice cloning.
+            progress_callback: Optional callback ``(stage: str, progress: float) -> None``.
+            transcription: Optional pre-computed ``Transcription`` to skip the Whisper step.
+
+        Returns:
+            ``DubbingResult`` with the dubbed audio, translated segments, and
+            source transcription. The output video is written to ``output_path``.
+        """
+        from videopython.ai.dubbing.remux import replace_audio_stream
+        from videopython.base.audio import Audio
+
+        input_path = Path(input_path)
+        output_path = Path(output_path)
+
+        if not input_path.exists():
+            raise FileNotFoundError(f"Input video not found: {input_path}")
+
+        logger.info("dub_file: loading audio from %s", input_path)
+        source_audio = Audio.from_path(input_path)
+
+        if self._local_pipeline is None:
+            self._init_local_pipeline()
+
+        result = self._local_pipeline.process(
+            source_audio=source_audio,
+            target_lang=target_lang,
+            source_lang=source_lang,
+            preserve_background=preserve_background,
+            voice_clone=voice_clone,
+            enable_diarization=enable_diarization,
+            progress_callback=progress_callback,
+            transcription=transcription,
+        )
+
+        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
+            dubbed_audio_path = Path(tmp.name)
+        try:
+            result.dubbed_audio.save(dubbed_audio_path)
+            replace_audio_stream(
+                video_path=input_path,
+                audio_path=dubbed_audio_path,
+                output_path=output_path,
+            )
+        finally:
+            dubbed_audio_path.unlink(missing_ok=True)
+
+        return result
+
     def revoice(
         self,
         video: Video,
@@ -101,7 +191,7 @@ class VideoDubber:
             self._init_local_pipeline()
 
         return self._local_pipeline.revoice(
-            video=video,
+            source_audio=video.audio,
             text=text,
             preserve_background=preserve_background,
             progress_callback=progress_callback,

{videopython-0.26.1 → videopython-0.26.3}/src/videopython/ai/dubbing/pipeline.py

@@ -9,18 +9,29 @@ from videopython.ai.dubbing.models import DubbingResult, RevoiceResult, Separate
 from videopython.ai.dubbing.timing import TimingSynchronizer
 
 if TYPE_CHECKING:
-    from videopython.base.video import Video
+    from videopython.base.audio import Audio
 
 logger = logging.getLogger(__name__)
 
 
 class LocalDubbingPipeline:
-    """Local pipeline for video dubbing."""
+    """Local pipeline for video dubbing.
 
-    def __init__(self, device: str | None = None):
+    When ``low_memory=True``, each stage's model is unloaded after it runs, so
+    only one model is resident at a time. This trades per-run latency (models
+    re-load from disk between stages) for peak memory. Recommended for GPUs
+    with <=12GB VRAM or hosts with <32GB RAM.
+    """
+
+    def __init__(self, device: str | None = None, low_memory: bool = False):
         self.device = device
+        self.low_memory = low_memory
         requested = device.lower() if isinstance(device, str) else "auto"
-        logger.info("LocalDubbingPipeline initialized with device=%s", requested)
+        logger.info(
+            "LocalDubbingPipeline initialized with device=%s low_memory=%s",
+            requested,
+            low_memory,
+        )
 
         self._transcriber: Any = None
         self._transcriber_diarization: bool | None = None
@@ -31,6 +42,23 @@ class LocalDubbingPipeline:
         self._separator: Any = None
         self._synchronizer: TimingSynchronizer | None = None
 
+    def _maybe_unload(self, component_name: str) -> None:
+        """Unload a stage's model when low_memory mode is enabled.
+
+        No-op when low_memory=False or the component was never initialized
+        (e.g. caller supplied a pre-computed transcription so the transcriber
+        was skipped).
+        """
+        if not self.low_memory:
+            return
+        component = getattr(self, component_name, None)
+        if component is None:
+            return
+        unload = getattr(component, "unload", None)
+        if callable(unload):
+            logger.info("low_memory: unloading %s", component_name.lstrip("_"))
+            unload()
+
     def _init_transcriber(self, enable_diarization: bool = False) -> None:
         """Initialize the transcription model."""
         from videopython.ai.understanding.audio import AudioToText
@@ -74,7 +102,6 @@ class LocalDubbingPipeline:
         max_duration: float = 10.0,
     ) -> dict[str, Any]:
         """Extract voice samples for each speaker from the audio."""
-        from videopython.base.audio import Audio
 
         voice_samples: dict[str, Audio] = {}
 
@@ -107,7 +134,7 @@ class LocalDubbingPipeline:
 
     def process(
         self,
-        video: Video,
+        source_audio: Audio,
         target_lang: str,
         source_lang: str | None = None,
         preserve_background: bool = True,
@@ -116,22 +143,22 @@ class LocalDubbingPipeline:
         progress_callback: Callable[[str, float], None] | None = None,
         transcription: Any | None = None,
     ) -> DubbingResult:
-        """Process a video through the local dubbing pipeline.
+        """Run the dubbing pipeline against the given source audio.
 
         Args:
+            source_audio: Source audio track to dub. Callers with a ``Video``
+                object should pass ``video.audio``; callers with only a file path
+                can use ``Audio.from_path(path)`` to avoid loading video frames.
             transcription: Optional pre-computed Transcription object. When provided,
                 the internal Whisper transcription step is skipped (saving time and VRAM).
                 Must be a ``videopython.base.text.transcription.Transcription`` instance
                 with populated ``segments``.
         """
-        from videopython.base.audio import Audio
 
         def report_progress(stage: str, progress: float) -> None:
             if progress_callback:
                 progress_callback(stage, progress)
 
-        source_audio = video.audio
-
         if transcription is not None:
             report_progress("Using provided transcription", 0.05)
         else:
@@ -141,6 +168,7 @@ class LocalDubbingPipeline:
                 self._transcriber_diarization = enable_diarization
 
             transcription = self._transcriber.transcribe(source_audio)
+            self._maybe_unload("_transcriber")
 
         if not transcription.segments:
             return DubbingResult(
@@ -162,6 +190,7 @@ class LocalDubbingPipeline:
                 self._init_separator()
 
             separated_audio = self._separator.separate(source_audio)
+            self._maybe_unload("_separator")
             vocal_audio = separated_audio.vocals
 
         voice_samples: dict[str, Audio] = {}
@@ -178,6 +207,7 @@ class LocalDubbingPipeline:
             target_lang=target_lang,
             source_lang=detected_lang,
         )
+        self._maybe_unload("_translator")
 
         report_progress("Generating dubbed speech", 0.50)
         if self._tts is None or self._tts_voice_clone != voice_clone or self._tts_language != target_lang:
@@ -208,6 +238,8 @@ class LocalDubbingPipeline:
             target_durations.append(segment.duration)
             start_times.append(segment.start)
 
+        self._maybe_unload("_tts")
+
         report_progress("Synchronizing timing", 0.85)
         if self._synchronizer is None:
             self._init_synchronizer()
@@ -242,19 +274,23 @@ class LocalDubbingPipeline:
 
     def revoice(
         self,
-        video: Video,
+        source_audio: Audio,
         text: str,
         preserve_background: bool = True,
         progress_callback: Callable[[str, float], None] | None = None,
     ) -> RevoiceResult:
-        """Replace speech in a video with new text using voice cloning."""
+        """Replace speech in audio with new text using voice cloning.
+
+        Args:
+            source_audio: Source audio track to revoice. Callers with a ``Video``
+                object should pass ``video.audio``.
+        """
         from videopython.base.audio import Audio
 
         def report_progress(stage: str, progress: float) -> None:
             if progress_callback:
                 progress_callback(stage, progress)
 
-        source_audio = video.audio
         original_duration = source_audio.metadata.duration_seconds
 
         report_progress("Analyzing audio", 0.05)
@@ -263,6 +299,7 @@ class LocalDubbingPipeline:
             self._transcriber_diarization = False
 
         transcription = self._transcriber.transcribe(source_audio)
+        self._maybe_unload("_transcriber")
 
         separated_audio: SeparatedAudio | None = None
         vocal_audio = source_audio
@@ -273,6 +310,7 @@ class LocalDubbingPipeline:
                 self._init_separator()
 
             separated_audio = self._separator.separate(source_audio)
+            self._maybe_unload("_separator")
             vocal_audio = separated_audio.vocals
 
         report_progress("Extracting voice sample", 0.40)
@@ -295,6 +333,7 @@ class LocalDubbingPipeline:
 
         generated_speech = self._tts.generate_audio(text, voice_sample=voice_sample)
         speech_duration = generated_speech.metadata.duration_seconds
+        self._maybe_unload("_tts")
 
         report_progress("Assembling audio", 0.85)
 

videopython-0.26.3/src/videopython/ai/dubbing/remux.py

@@ -0,0 +1,73 @@
+"""ffmpeg helper for replacing a video file's audio track without re-encoding video."""
+
+from __future__ import annotations
+
+import logging
+import subprocess
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+
+class RemuxError(RuntimeError):
+    """ffmpeg failed while replacing an audio stream."""
+
+
+def replace_audio_stream(
+    video_path: str | Path,
+    audio_path: str | Path,
+    output_path: str | Path,
+    audio_codec: str = "aac",
+    audio_bitrate: str = "192k",
+) -> None:
+    """Copy ``video_path``'s video stream and mux in ``audio_path`` as the audio track.
+
+    Uses ffmpeg stream-copy for video (no re-encode) and encodes audio to AAC.
+    ``-shortest`` trims to the shorter of the two streams so the output duration
+    matches the source video when the dubbed audio is slightly longer.
+
+    Args:
+        video_path: Source video file (video stream is copied unchanged).
+        audio_path: Audio file to use as the new audio track.
+        output_path: Destination file. Overwritten if it exists.
+        audio_codec: ffmpeg audio codec name. Defaults to ``aac`` (MP4-compatible).
+        audio_bitrate: Audio bitrate passed to ffmpeg (``-b:a``).
+
+    Raises:
+        FileNotFoundError: If ``video_path`` or ``audio_path`` does not exist.
+        RemuxError: If ffmpeg returns a non-zero exit code.
+    """
+    video_path = Path(video_path)
+    audio_path = Path(audio_path)
+    output_path = Path(output_path)
+
+    if not video_path.exists():
+        raise FileNotFoundError(f"Video file not found: {video_path}")
+    if not audio_path.exists():
+        raise FileNotFoundError(f"Audio file not found: {audio_path}")
+
+    cmd = [
+        "ffmpeg",
+        "-y",
+        "-i",
+        str(video_path),
+        "-i",
+        str(audio_path),
+        "-map",
+        "0:v:0",
+        "-map",
+        "1:a:0",
+        "-c:v",
+        "copy",
+        "-c:a",
+        audio_codec,
+        "-b:a",
+        audio_bitrate,
+        "-shortest",
+        str(output_path),
+    ]
+
+    logger.info("replace_audio_stream: %s + %s -> %s", video_path, audio_path, output_path)
+    result = subprocess.run(cmd, capture_output=True)
+    if result.returncode != 0:
+        raise RemuxError(f"ffmpeg failed (exit {result.returncode}): {result.stderr.decode(errors='replace')}")

{videopython-0.26.1 → videopython-0.26.3}/src/videopython/ai/generation/audio.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from typing import Any
 
-from videopython.ai._device import log_device_initialization, select_device
+from videopython.ai._device import log_device_initialization, release_device_memory, select_device
 from videopython.base.audio import Audio, AudioMetadata
 
 
@@ -151,6 +151,16 @@ class TextToSpeech:
 
         return self._generate_local(text, effective_voice)
 
+    def unload(self) -> None:
+        """Release the TTS model(s) so the next generate_audio() re-initializes.
+
+        Used by low-memory dubbing to free VRAM between pipeline stages.
+        """
+        self._model = None
+        self._processor = None
+        self._chatterbox_model = None
+        release_device_memory(self.device)
+
 
 class TextToMusic:
     """Generates music from text descriptions using MusicGen."""

{videopython-0.26.1 → videopython-0.26.3}/src/videopython/ai/generation/translation.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from typing import Any
 
-from videopython.ai._device import log_device_initialization, select_device
+from videopython.ai._device import log_device_initialization, release_device_memory, select_device
 from videopython.ai.dubbing.models import TranslatedSegment
 from videopython.base.text.transcription import TranscriptionSegment
 
@@ -180,6 +180,16 @@ class TextTranslator:
 
         return translated_segments
 
+    def unload(self) -> None:
+        """Release the translation model so the next translate() re-initializes.
+
+        Used by low-memory dubbing to free VRAM between pipeline stages.
+        """
+        self._model = None
+        self._tokenizer = None
+        self._current_lang_pair = None
+        release_device_memory(self.device)
+
     @staticmethod
     def get_supported_languages() -> dict[str, str]:
         return LANGUAGE_NAMES.copy()

{videopython-0.26.1 → videopython-0.26.3}/src/videopython/ai/understanding/audio.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from typing import Any, Literal
 
-from videopython.ai._device import log_device_initialization, select_device
+from videopython.ai._device import log_device_initialization, release_device_memory, select_device
 from videopython.base.audio import Audio
 from videopython.base.description import AudioClassification, AudioEvent
 from videopython.base.text.transcription import Transcription, TranscriptionSegment, TranscriptionWord
@@ -51,6 +51,15 @@ class AudioToText:
         self._diarization_pipeline = Pipeline.from_pretrained(self.PYANNOTE_DIARIZATION_MODEL)
         self._diarization_pipeline.to(torch.device(self.device))
 
+    def unload(self) -> None:
+        """Release the Whisper and diarization models so the next call re-initializes.
+
+        Used by low-memory dubbing to free VRAM between pipeline stages.
+        """
+        self._model = None
+        self._diarization_pipeline = None
+        release_device_memory(self.device)
+
     def _process_transcription_result(self, transcription_result: dict) -> Transcription:
         """Process raw transcription result into a Transcription object."""
         transcription_segments = []

{videopython-0.26.1 → videopython-0.26.3}/src/videopython/ai/understanding/separation.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from typing import Any
 
-from videopython.ai._device import log_device_initialization, select_device
+from videopython.ai._device import log_device_initialization, release_device_memory, select_device
 from videopython.ai.dubbing.models import SeparatedAudio
 from videopython.base.audio import Audio, AudioMetadata
 
@@ -134,3 +134,11 @@ class AudioSeparator:
     def extract_background(self, audio: Audio) -> Audio:
         """Convenience method to extract only background from audio."""
         return self.separate(audio).background
+
+    def unload(self) -> None:
+        """Release the Demucs model so the next separate() re-initializes.
+
+        Used by low-memory dubbing to free VRAM between pipeline stages.
+        """
+        self._model = None
+        release_device_memory(self.device)