webm video presets: migrate from VP8 to VP9 for smaller ZIMs

WebmLow and WebmHigh presets are migrated to use VP9 instead of VP8.

Presets are also enhanced to maximize video quality all while minimizing
final file size. A compromise with encoding duration (which is known to
be significantly higher with VP9 compared to VP8) is also taken into
account (hence the -speed setting values, 0 is optimal but takes too
long for minimal quality improvement given our computing power available)

CHANGELOG.md

@@ -8,6 +8,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Changed
+
+- Migrate the **VideoWebmLow** and **VideoWebmHigh** presets to VP9 for smaller file size #79
+  - New preset versions are v3 and v2 respectively
 - Simplify type annotations by replacing Union and Optional with pipe character ("|") for improved readability and clarity
 
 ### Fixed

src/zimscraperlib/video/presets.py

@@ -37,21 +37,23 @@ class VideoWebmLow(Config):
     128k target video bitrate but stay within quality boundaries.
     48k audio bitrate"""
 
-    VERSION = 2
+    VERSION = 3
 
     ext = "webm"
     mimetype = f"{preset_type}/webm"
 
     options: ClassVar[dict[str, str | bool | int | None]] = {
-        "-codec:v": "libvpx",  # video codec
-        "-quality": "best",  # codec preset
-        "-b:v": "128k",  # Adjust quantizer within min/max to target this bitrate
-        "-qmin": "18",  # Reduce the bitrate on very still videos
+        "-codec:v": "libvpx-vp9",  # video codec
+        "-b:v": "140k",  # Adjust quantizer within min/max to target this bitrate
+        "-qmin": "30",  # Reduce the bitrate on very still videos
         "-qmax": "40",  # Increase the bitrate on very busy videos
+        "-g": "240",  # Number of frames allowed between keyframes
+        "-quality": "good",  # codec preset
+        "-speed": "4",  # Encoding speed (compromise between quality and encoding time)
         "-vf": "scale='480:trunc(ow/a/2)*2'",  # frame size
         "-codec:a": "libvorbis",  # audio codec
-        "-ar": "44100",  # audio sampling rate
         "-b:a": "48k",  # target audio bitrate
+        "-ar": "44100",  # audio sampling rate
     }
 
 
@@ -88,16 +90,22 @@ class VideoWebmHigh(Config):
 
     25 constant quality"""
 
-    VERSION = 1
+    VERSION = 2
 
     ext = "webm"
     mimetype = f"{preset_type}/webm"
 
     options: ClassVar[dict[str, str | bool | int | None]] = {
-        "-codec:v": "libvpx",  # video codec
+        "-codec:v": "libvpx-vp9",  # video codec
+        "-b:v": "340k",  # Adjust quantizer within min/max to target this bitrate
+        "-qmin": "26",  # Reduce the bitrate on very still videos
+        "-qmax": "54",  # Increase the bitrate on very busy videos
+        "-g": "240",  # Number of frames allowed between keyframes
+        "-quality": "good",  # codec preset
+        "-speed": "1",  # Encoding speed (compromise between quality and encoding time)
         "-codec:a": "libvorbis",  # audio codec
-        "-crf": "25",  # constant quality, lower value gives better qual and larger size
-        "-b:v": "0",  # must be passed if using constant quality mode via -cbr for codec
+        "-b:a": "48k",  # target audio bitrate
+        "-ar": "44100",  # audio sampling rate
     }
 
 

tests/video/test_video.py

@@ -151,19 +151,21 @@ def test_preset_has_mime_and_ext():
 
 def test_preset_video_webm_low():
     config = VideoWebmLow()
-    assert config.VERSION == 2
+    assert config.VERSION == 3
     args = config.to_ffmpeg_args()
-    assert len(args) == 20
+    assert len(args) == 24
     options_map = [
-        ("codec:v", "libvpx"),
+        ("codec:v", "libvpx-vp9"),
         ("codec:a", "libvorbis"),
-        ("b:v", "128k"),
+        ("b:v", "140k"),
         ("ar", "44100"),
         ("b:a", "48k"),
-        ("quality", "best"),
-        ("qmin", "18"),
+        ("quality", "good"),
+        ("qmin", "30"),
         ("qmax", "40"),
         ("vf", "scale='480:trunc(ow/a/2)*2'"),
+        ("g", "240"),
+        ("speed", "4"),
     ]
     for option, val in options_map:
         idx = args.index(f"-{option}")
@@ -172,35 +174,41 @@ def test_preset_video_webm_low():
 
     # test updating values
     config = VideoWebmLow(**{"-ar": "50000"})
-    config["-bufsize"] = "900k"
+    config["-qmin"] = "25"
     args = config.to_ffmpeg_args()
     idx = args.index("-ar")
     assert idx != -1 and args[idx + 1] == "50000"
-    idx = args.index("-bufsize")
-    assert idx != -1 and args[idx + 1] == "900k"
+    idx = args.index("-qmin")
+    assert idx != -1 and args[idx + 1] == "25"
 
 
 def test_preset_video_webm_high():
     config = VideoWebmHigh()
-    assert config.VERSION == 1
+    assert config.VERSION == 2
     args = config.to_ffmpeg_args()
-    assert len(args) == 10
+    assert len(args) == 22
     options_map = [
-        ("codec:v", "libvpx"),
+        ("codec:v", "libvpx-vp9"),
         ("codec:a", "libvorbis"),
-        ("b:v", "0"),
-        ("crf", "25"),
+        ("b:v", "340k"),
+        ("ar", "44100"),
+        ("b:a", "48k"),
+        ("quality", "good"),
+        ("qmin", "26"),
+        ("qmax", "54"),
+        ("g", "240"),
+        ("speed", "1"),
     ]
     for option, val in options_map:
         idx = args.index(f"-{option}")
         assert idx != -1
         assert args[idx + 1] == val
 
     # test updating values
-    config = VideoWebmHigh(**{"-crf": "51"})
+    config = VideoWebmHigh(**{"-qmax": "51"})
     config["-b:v"] = "300k"
     args = config.to_ffmpeg_args()
-    idx = args.index("-crf")
+    idx = args.index("-qmax")
     assert idx != -1 and args[idx + 1] == "51"
     idx = args.index("-b:v")
     assert idx != -1 and args[idx + 1] == "300k"
@@ -314,7 +322,7 @@ def test_preset_voice_mp3_low():
             "video.mp4",
             "video.webm",
             VideoWebmLow().to_ffmpeg_args(),
-            {"codecs": ["vp8", "vorbis"], "duration": 2},
+            {"codecs": ["vp9", "vorbis"], "duration": 2},
         ),
         (
             "video.webm",