[Regression] AAC audio in RTSP no more muxed with "-c copy"

[HelFix]

Summary of the bug:

• How to reproduce:
• Use the following command with an RTSP stream:

  ffmpeg -rtsp_transport tcp -i rtsp://admin:admin@10.15.129.19:554/Streaming/Channels/102 -c:v copy -c:a copy output.mp4
  

• Test with:

ffmpeg-n7.1-39-g64e2864cb9-linux64-gpl-7.1 or 6.1.1 (works).
ffmpeg-n7.1-62-gb168ed9b14-linux64-gpl-7.1 and newer (fails).

• Expected Behavior:

Both video and audio streams are muxed into output.mp4. In n7.1-39, ffprobe shows:

Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'output.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2avc1mp41
    title           : Media Presentation
    encoder         : Lavf61.7.100
  Duration: 00:00:04.95, start: 0.000000, bitrate: 2591 kb/s
  Stream #0:0[0x1](und): Video: h264 (Baseline) (avc1 / 0x31637661), yuvj420p(pc, progressive), 1920x536 [SAR 1:1 DAR 240:67], 2558 kb/s, 19.80 fps, 20 tbr, 90k tbn (default)
      Metadata:
        handler_name    : VideoHandler
        vendor_id       : [0][0][0][0]
  Stream #0:1[0x2](und): Audio: aac (LC) (mp4a / 0x6134706D), 32000 Hz, mono, fltp, 26 kb/s (default)
      Metadata:
        handler_name    : SoundHandler
        vendor_id       : [0][0][0][0]

• Actual Behavior:

In n7.1-62:

Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'output.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2avc1mp41
    title           : Media Presentation
    encoder         : Lavf61.7.100
  Duration: 00:00:09.15, start: 0.000000, bitrate: 1918 kb/s
  Stream #0:0[0x1](und): Video: h264 (Baseline) (avc1 / 0x31637661), yuvj420p(pc, progressive), 1920x536 [SAR 1:1 DAR 240:67], 1917 kb/s, 19.34 fps, 20 tbr, 90k tbn (default)
      Metadata:
        handler_name    : VideoHandler
        vendor_id       : [0][0][0][0]

Video is muxed (e.g., 219 packets, 2116297 bytes).
Audio is read (e.g., 366 packets, 39746 bytes) but not muxed (0 packets, 0 bytes).
Example log from a similar 7.1 failure:

[in#0/rtsp] Input stream #0:1 (audio): 366 packets read (39746 bytes);
[out#0/mp4] Output stream #0:1 (audio): 0 packets muxed (0 bytes);

[rtsp @ 0x6044978bfbc0] SDP:
v=0
o=- 1743073449707371 1743073449707371 IN IP4 10.15.129.19
s=Media Presentation
e=NONE
b=AS:5100
t=0 0
a=control:rtsp://10.15.129.19:554/Streaming/Channels/102/
m=video 0 RTP/AVP 96
c=IN IP4 0.0.0.0
b=AS:5000
a=recvonly
a=x-dimensions:1920,536
a=control:rtsp://10.15.129.19:554/Streaming/Channels/102/trackID=1
a=rtpmap:96 H264/90000
a=fmtp:96 profile-level-id=420029; packetization-mode=1; sprop-parameter-sets=Z0IAM4qKUDwCK8v/gACAALZAAAD6AAATiAE=,aM48gA==
m=audio 0 RTP/AVP 104
c=IN IP4 0.0.0.0
b=AS:50
a=recvonly
a=control:rtsp://10.15.129.19:554/Streaming/Channels/102/trackID=2
a=rtpmap:104 mpeg4-generic/32000/1
a=fmtp:104 profile-level-id=15; streamtype=5; mode=AAC-hbr; config=1288;SizeLength=13; IndexLength=3; IndexDeltaLength=3; Profile=1;
a=Media_header:MEDIAINFO=494D4B48010300000400000101200110007D0000007D000000000000000000000000000000000000;
a=appversion:1.0

• Environment:

OS: Ubuntu 24.04
Builds: BtbN static Linux 64-bit GPL from ​https://github.com/BtbN/FFmpeg-Builds.

Patches should be submitted to the ffmpeg-devel mailing list and not this bug tracker.

[MasterQuestionable]

͏    Maybe the audio is outright not even recognized..?

͏    Regression window:
͏    ​https://github.com/FFmpeg/FFmpeg/commits?since=2024-11-18&until=2024-12-20
͏    ​https://github.com/search?type=commits&q=repo:FFmpeg/FFmpeg+aac+committer-date:2024-11-18T20:09:07Z..2024-12-20T01:06:22Z&s=committer-date&o=asc

͏    @ 2024-11-24T23:14:11Z:
͏    ​avformat/movenc: write stss boxes for xHE-AAC
͏    @ 2024-11-24T23:26:24Z:
͏    ​avcodec/aac_parser: set key_frame and profile
͏    @ 2024-11-24T23:26:26Z:
͏    ​avcodec/aac/aacdec: set keyframe flag in output frames
͏    @ 2024-11-24T23:38:15Z:
͏    ​avcodec/codec_desc: remove Intra Only prop for AAC
͏    @ 2024-12-11T20:55:40Z:
͏    ​avcodec/aac/aacdec_usac: Clean ics2->max_sfb when first SCE fails
͏    @ 2024-12-11T20:56:13Z:
͏    ​avcodec/aac/aacdec_usac: Dont leave type at a invalid value

[HelFix]

i found that commit

The regression is caused by commit f265f9c9d04863180503707bfad285f48e6bf080

• avcodec/codec_desc: remove Intra Only prop for AAC

[MasterQuestionable]

͏    What's the input characteristics?
͏    Does it affect simply the AAC file?

[HelFix]

Replying to MasterQuestionable:

͏    What's the input characteristics?
͏    Does it affect simply the AAC file?

Hello, what do you mean, do you have any command what i can run? Or is there any way to record that rtsp raw?

[MasterQuestionable]

͏    The regressor commit doesn't look like RTSP specific.
͏    So I suspect it would also fail the plain AAC.

͏    Try "-c copy" the RTSP stream with a previous working FFmpeg.
͏    ("-map" only the audio and output as ".aac")
͏    .
͏    Test whether the AAC alone could reproduce the same problem.

[HelFix]

ok, i tried

ffmpeg version n7.1-55-g4b801baa46

ffmpeg -rtsp_transport tcp -i RTPS -c:v copy -c:a copy -map 0:a output.aac

ffmpeg version n7.1-56-gf265f9c9d0

ffmpeg -i output.aac -c:a copy -map 0:a output2.aac

i wasn't able to replicate same issue.

so i setup public restream from my rtsp for only audio

rtsp://185.91.117.52:8554/mystream
(fixed)

if i do

ffmpeg -rtsp_transport tcp -i rtsp://185.91.117.52:8554/mystream -c:a copy output.aac

ffmpeg version n7.1-55-g4b801baa46 - working ok
ffmpeg version n7.1-56-gf265f9c9d0 - output.aac output size is 0 kb

[Jack Lau]

Replying to HelFix:

ffmpeg version n7.1-56-gf265f9c9d0 - output.aac output size is 0 kb

I think it's not a bug, you need to add option -copyinkf in your FFmpeg command if the aac not have k frames.

ffmpeg -rtsp_transport tcp -i rtsp://localhost:8554/cap/live -c:a copy -copyinkf test.aac -y

In my case, it works well.

[Jack Lau]

i found the root reason after careful research，

this indeed a bug because rtsp aac didn't set the need_parsing so that it cannot mark keyframe

Should be fixed in this patch https://ffmpeg.org/pipermail/ffmpeg-devel/2025-April/341989.html

[Jack Lau]

pushed in cf7dfddee4068c65811965dcea42d9a86c104bca