Twitter MP3 vs M4A: Choose the Source Before 320kbps

The same post can offer different audio tracks

On October 15, 2026, the public metadata returned five direct MP4 variants for the SpaceX clip and three for the NASA Artemis clip. Each file was downloaded, inspected with ffprobe, and hashed twice: once as its AAC packet payload and once after decoding to signed 16-bit PCM. All eight tracks were AAC LC at 48 kHz.

The resolution below identifies the MP4 file. Its advertised video bitrate is not the audio bitrate; the audio figures come from that file’s audio stream.

1. SpaceX: 480 × 270 held 32.26 kb/s mono; 640 × 360 held 64.45 kb/s stereo; 720p, 1080p and 4K held 128.06 kb/s stereo. The three highest-resolution files had identical audio packet and decoded-PCM hashes.
2. For that SpaceX post, the 720p MP4 was 1,446,945 bytes; the 4K file was 21,691,445 bytes. Downloading 4K added no different audio payload in this comparison.
3. NASA Artemis: the three returned MP4s held 33.03, 64.50 and 128.07 kb/s stereo audio. Their packet and PCM hashes differed. The highest measured audio rate was in the 720p file.

Choose the MP4 input before pressing an audio button

CarryFeed’s M4A, MP3 128k and MP3 320k buttons all use the same automatically selected input. The current picker sorts direct MP4 options by their advertised video bitrate, lowest first. Choosing 320k changes the output encoding; it does not switch to another source MP4.

To use a different track, open the video downloader, save the MP4 option you want to inspect, then run the local commands below. The Video source options on the audio page also download MP4 files; clicking one does not change the audio buttons’ input.

For these two posts, the 720p files supplied the highest measured AAC rate. On the SpaceX clip, 720p also matched the 1080p and 4K audio exactly. On another post, inspect its own tracks rather than selecting by resolution alone.

The eight-variant measurement CSV includes source URLs, stream fields, byte counts and full hashes. This pass compared direct MP4 files; it did not examine HLS playlists or measure listening preference. A hash comparison establishes equality of the tested payloads or decoded output, not a subjective sound-quality score.

After source selection: one June clip, three output files

In the June 20 check, the FORAB clip lasted 20.31 seconds. Its selected source carried mono AAC LC at 48 kHz and 32 kb/s. The measurements below use that same track for each conversion.

The 12 historical output-size measurements

The checks resolved public post media, selected the lowest-bitrate direct MP4 using the audio UI’s source preference, inspected that file with ffmpeg, and created M4A copy, MP3 128k and MP3 320k outputs.

All 12 selected tracks were AAC LC at 31–33 kb/s: eight at 48 kHz and four at 44.1 kHz. These describe the automatically selected low-video-bitrate MP4s. The October comparison above shows why they should not stand in for every track available from the same posts.

The mean outputs were 0.24 MiB per minute for M4A, 0.92 for MP3 128k and 2.31 for MP3 320k. The near-2.5× ratio between MP3 outputs follows the requested 320/128 bitrate ratio; it is a storage tradeoff, not a listening result.

Choose for the destination

For storage or an app that accepts M4A, copying the AAC stream avoids another lossy encoding step. Open the resulting file in the destination app once to confirm it works.

For an MP3-only upload, use the required MP3 setting. If no bitrate is specified, 128k is the smaller option shown here. Check playback before handing it off.

Use 320k when an app or publishing specification requires it. Converting a low-bitrate source to a larger MP3 cannot reconstruct sound absent from the source.

The three silent checks stopped before conversion

Three GIF-style MP4s had video streams and no audio stream. The MP3 attempts ended with “Output file does not contain any stream.” There was no track to encode.

1. Check the original video for sound and inspect the selected MP4 for an Audio stream.
2. For a silent looping clip, keep the GIF-style video rather than retrying another audio bitrate.
3. If the source has sound but the saved file is silent, use the no-sound checks.

Repeat the check on one source

Save the chosen MP4 as source.mp4. Inspect its audio stream, then copy AAC into M4A or encode the MP3 rate the destination requires. Use another local filename when comparing a second MP4 from the same post.

Keep the original post URL, MP4 variant URL, audio stream fields and output filename in one note. The video-option bitrate is not an audio measurement. For an identity comparison, run the final hash command on each local input.

ffprobe -v error -select_streams a:0 -show_entries stream=codec_name,profile,sample_rate,channels,bit_rate,duration -of json source.mp4

ffmpeg -y -i source.mp4 -vn -c:a copy output.m4a

ffmpeg -y -i source.mp4 -vn -c:a libmp3lame -b:a 128k output-128k.mp3

ffmpeg -y -i source.mp4 -vn -c:a libmp3lame -b:a 320k output-320k.mp3

ffmpeg -v error -i source.mp4 -map 0:a:0 -c:a pcm_s16le -f hash -hash sha256 -