[nicovideo.jp] Extracting the audio from nicovideos in the "aac" or "m4a" file format causes the length to be missing, but NOT "alac"

[Synth-ix]

[nicovideo.jp] Extracting the audio from nicovideos in the "aac" or "m4a" file format causes the length to be missing, but NOT "alac"

[bashonly]

Niconico is known to only have MP4 format (according to this and this Q&A pages from the niconico official site), that's why it was hard-coded to m4a.

Right, but those such cases are typically handled by passing mp4 as the ext param to _extract_m3u8_formats(), which the above patch does

[bashonly]

from the m4a log:

[ExtractAudio] Not converting audio 【マリオペイント】お猿のおもちゃのBGMをリミックスしてみた [nm10550338].m4a; the file is already in a common audio format

Here, the only thing that -x is doing is implicitly selecting the bestaudio format. yt-dlp detects that the download is already in an audio-only container (m4a) and does not extract or remux anything. So you end up with the raw product of the m3u8 playlist; the native HLS downloader just concatenates all of the AAC audio stream segments. The resulting container is not playing nice with Windows' metadata parsing.

When you pass --audio-format alac (or any format other than m4a or aac), then yt-dlp will remux or re-encode the download, which effectively fixes the issue.

Possible solutions/workarounds for getting a fixed m4a container:

1. use -x -f bv+ba/b to download the video and the audio, merge them, and then extract the audio
2. use -x --parse-metadata " m4a_dash: %(container)s" to download only audio and force remuxing after download
3. use ffmpeg as the downloader: --downloader ffmpeg
4. manually remux with ffmpeg after downloading, e.g.: ffmpeg -i "input.m4a" -c copy "output.m4a"

[Synth-ix]

Those solutions worked just perfectly!

Thank you for the explanation & inputs on fixing the issue, seriously. ♥

[bashonly]

Looking at the extractor code though, I'm noticing we manually set the audio format extension to m4a. I'm wondering if this should be fixed on the extractor level like this:

diff --git a/yt_dlp/extractor/niconico.py b/yt_dlp/extractor/niconico.py
index 179e7a9b1..e06740d62 100644
--- a/yt_dlp/extractor/niconico.py
+++ b/yt_dlp/extractor/niconico.py
@@ -420,7 +420,7 @@ def _yield_dms_formats(self, api_data, video_id):
                 'x-request-with': 'https://www.nicovideo.jp',
             })['data']['contentUrl']
         # Getting all audio formats results in duplicate video formats which we filter out later
-        dms_fmts = self._extract_m3u8_formats(dms_m3u8_url, video_id)
+        dms_fmts = self._extract_m3u8_formats(dms_m3u8_url, video_id, 'mp4')
 
         # m3u8 extraction does not provide audio bitrates, so extract from the API data and fix
         for audio_fmt in traverse_obj(dms_fmts, lambda _, v: v['vcodec'] == 'none'):
@@ -432,7 +432,6 @@ def _yield_dms_formats(self, api_data, video_id):
                     'asr': ('samplingRate', {int_or_none}),
                 }), get_all=False),
                 'acodec': 'aac',
-                'ext': 'm4a',
             }
 
         # Sort before removing dupes to keep the format dicts with the lowest tbr

[Synth-ix]

Definitely shoot me a reply if/when you'd go about releasing a build including this potential patch, I'd be willing to test it out before release if for whatever reason that'd be needed as well.
Glad I was able to bring this to light & you're able to find a possible way to go about troubleshooting this!

[CXwudi]

Niconico is known to only have MP4 format (according to this and this Q&A pages from the niconico official site), that's why it was hard-coded to m4a.