✨(summary) handle video files more efficiently

Video files are heavy recording files, sometimes several hours long.

Previously, recordings were naively submitted to the Whisper API without
chunking, resulting in very large requests that could take a long time
to process. Video files are much larger than audio-only files, which
could cause performance issues during upload.

Introduce an extra step to extract the audio component from MP4 files,
producing a lighter audio-only file (to be confirmed). No re-encoding
is done, just a minimal FFmpeg extraction based on community guidance,
since I’m not an FFmpeg expert.

This feature is experimental and may introduce regressions, especially
if audio quality or sampling is impacted, which could reduce Whisper’s
accuracy. Early tests with the ASR model worked, but it has not been
tested on long recordings (e.g., 3-hour meetings),
which some users have.

src/summary/Dockerfile

@@ -1,5 +1,13 @@
 FROM python:3.13-slim AS base
 
+
+# Install ffmpeg for audio/video processing (format conversion, extraction, compression)
+# See summary/core/file_service.py for usage.
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends ffmpeg=7:7.1.3-* && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
 FROM base AS builder
 
 WORKDIR /app

src/summary/summary/core/config.py

@@ -20,6 +20,7 @@ class Settings(BaseSettings):
     # Audio recordings
     recording_max_duration: Optional[int] = None
     recording_allowed_extensions: Set[str] = {".ogg", ".mp4"}
+    recording_video_extensions: Set[str] = {".mp4"}
 
     # Celery settings
     celery_broker_url: str = "redis://redis/0"

src/summary/summary/core/file_service.py

@@ -1,6 +1,7 @@
 """File service to encapsulate files' manipulations."""
 
 import os
+import subprocess
 import tempfile
 from contextlib import contextmanager
 from pathlib import Path
@@ -95,6 +96,46 @@ class FileService:
 
         return duration
 
+    def _extract_audio_from_video(self, video_path: Path) -> Path:
+        """Extract audio from video file (e.g., MP4) and save as audio file."""
+        self._logger.info("Extracting audio from video file: %s", video_path)
+
+        with tempfile.NamedTemporaryFile(
+            suffix=".m4a", delete=False, prefix="audio_extract_"
+        ) as tmp:
+            output_path = Path(tmp.name)
+
+        try:
+            command = [
+                "ffmpeg",
+                "-i",
+                str(video_path),
+                "-vn",  # No video
+                "-acodec",
+                "copy",
+                "-y",  # Overwrite output file if exists
+                str(output_path),
+            ]
+
+            # ruff: noqa: S603
+            subprocess.run(
+                command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=True
+            )
+
+            self._logger.info("Audio successfully extracted to: %s", output_path)
+            return output_path
+
+        except FileNotFoundError as e:
+            self._logger.error("ffmpeg not found. Please install ffmpeg.")
+            if output_path.exists():
+                os.remove(output_path)
+            raise RuntimeError("ffmpeg is not installed or not in PATH") from e
+        except subprocess.CalledProcessError as e:
+            self._logger.error("Audio extraction failed: %s", e.stderr.decode())
+            if output_path.exists():
+                os.remove(output_path)
+            raise RuntimeError("Failed to extract audio.") from e
+
     @contextmanager
     def prepare_audio_file(self, remote_object_key: str):
         """Download and prepare audio file for processing.
@@ -104,6 +145,7 @@ class FileService:
         when the context exits.
         """
         downloaded_path = None
+        processed_path = None
         file_handle = None
 
         try:
@@ -111,16 +153,24 @@ class FileService:
             duration = self._validate_duration(downloaded_path)
 
             extension = downloaded_path.suffix.lower()
+
+            if extension in settings.recording_video_extensions:
+                self._logger.info("Video file detected, extracting audio...")
+                extracted_audio_path = self._extract_audio_from_video(downloaded_path)
+                processed_path = extracted_audio_path
+            else:
+                processed_path = downloaded_path
+
             metadata = {"duration": duration, "extension": extension}
 
-            file_handle = open(downloaded_path, "rb")
+            file_handle = open(processed_path, "rb")
             yield file_handle, metadata
 
         finally:
             if file_handle:
                 file_handle.close()
 
-            for path in [downloaded_path]:
+            for path in [downloaded_path, processed_path]:
                 if path is None or not os.path.exists(path):
                     continue