Skip to content
Language:DeutschEnglish Last updated: 2026-10-23

Audio and video: formats and significant properties

This page sets out, for sound and video recordings, which formats Anton prefers and accepts, which properties must be preserved, and what Anton measures or checks of them today. It belongs to the series per format category (anton#687), after images, PDF and text. The choices are provisional: they are being tried in practice and will be revised where they do not prove themselves.

Anton needs this for three things: to document what makes up a recording (NDSA Cn-1.1), to verify that it still does (Cn-2.1), and to prove after a migration that nothing significant was lost (media:migrate).

Formats

Audio

Level Format Condition
preferred WAV or Broadcast Wave (BWF) with LPCM
preferred FLAC
accepted MP3, AAC (M4A), AIFF, Ogg Vorbis, Opus, WMA as long as ffmpeg reads them

Video

Level Format Condition
preferred original codec in a common container, as delivered born-digital video
preferred MKV with FFV1 and FLAC digitisation of analogue carriers
preferred MP4 with H.264 where storage is scarce
accepted MOV (ProRes too), AVI, MPEG-2, MPEG-TS, MXF, WebM, WMV, MP4 with HEVC or AV1 as long as ffmpeg reads them

Preferred for sound are the uncompressed or losslessly compressed formats WAV/BWF and FLAC; this is what NARA and KOST recommend. For video the codec comes first: a born-digital video is not re-encoded, since every lossy re-encoding loses information and may produce artefacts (KOST). When analogue carriers are digitised, the master is made losslessly, as MKV with FFV1 for the picture and FLAC for the sound (KOST); for sound, Memoriav and the Swiss National Sound Archives recommend 96 kHz and 24 bits, NARA at least 44.1 kHz and 16 bits, better 96 kHz and 24 bits, and 10 bits for video. MP4 with H.264 is the choice where a lossless master needs too much storage; an appraisal decision, as with JPEG for images.

Accepted is, as with images, what widespread open-source tools can read and will in all likelihood go on reading; for sound and video that is ffmpeg, which also decodes ProRes, WMV and MXF. Recordings already compressed with loss (MP3, AAC, H.264) are not converted to a lossless format: the loss remains, the file only grows (KOST). NARA advises converting WMV to AVI; under the principle it stays until ffmpeg no longer reads it.

Changing the container is possible without loss: the streams are moved unchanged into another container, for instance from MOV to MKV (KOST). This is the gentlest form of normalisation.

Here too: an archive keeps as few formats as possible.

Significant properties

Grouped as NARA does into appearance, structure, behaviour and context. NARA requires that a conversion of born-digital material keep the same sampling attributes, the same channel layout and, where the new format allows, the same compression.

Group Property Applies to Source
Appearance width and height in pixels, aspect ratio video NARA
frame rate, interlacing video NARA
bit depth and chroma subsampling video NARA
colour space video NARA
Structure duration audio, video NARA
number and kind of streams (picture, sound, subtitles) audio, video NARA
channels and channel layout (mono, stereo, 5.1) audio, video NARA
sample rate and bit depth audio, video NARA
content of each stream, where the format is lossless audio, video Anton
codec and compression: recorded, not preserved audio, video NARA, KOST
Behaviour The recording can be played, picture and sound in sync. audio, video NARA
Context embedded metadata (title, creator, recording date, for BWF the coding history) audio, video NARA

Codec and compression are not properties that must be preserved. They are recorded because they tell whether the other properties are held losslessly; and a migration should change them only where necessary.

For lossless formats, the content of a stream can be proved with a checksum per stream, the counterpart of the pixel signature for images: if it stays the same after a change of container or a lossless conversion, no frame and no sample has changed.

What Anton measures and checks

As of version 0.111.1.

Property Measured Checked
Container, duration, number of video, audio, subtitle and data streams with ffprobe at ingest; for older recordings media:validate-formats fills them in no
Picture: codec, width and height, aspect ratio (SAR, DAR), frame rate, interlacing, pixel format and bit depth, colour space likewise, for the first video stream; a cover picture does not count as a video stream no
Sound: codec, channels and channel layout, sample rate, bit depth likewise, for the first audio stream no
All streams likewise, one line per stream with kind and codec no
Content of each stream no (anton#717) no
Playable – at ingest ffprobe reads the container; nightly with media:validate-formats ffmpeg decodes every frame and every sample, and every error message counts; also when the access copy is made
Embedded metadata no no

Measured values appear on the medium under Technical characteristics. The older details on duration, codec, bit rate, resolution, sample rate and aspect ratio remain alongside for the Memobase and RiC exports; media:extract-av-metadata fills them.

Limits

  • Anton measures detailed characteristics only for the first video and the first audio stream; further streams are only counted and listed with kind and codec.
  • On umberto, on 2026-10-23, 769 of 771 sound and video recordings lacked the older details for the exports; media:extract-av-metadata has never run there.
  • media:migrate proves nothing for sound and video; the pixel signature exists only for images, a checksum per stream is missing (anton#717). There is no rule yet that converts sound or video.
  • Access copies are made as MP3 for sound and as MP4 with H.264, poster and thumbnail for video, from all formats above, FFV1 and ProRes included (anton#718). Upload takes them and keeps their extension.
  • «Playable» means ffmpeg decodes the recording without an error message. It does not check whether picture and sound are good.

What a migration must check

When media:migrate converts a recording, afterwards:

  • Duration, number and kind of streams, channels, sample rate, resolution and frame rate are the same.
  • With a lossless target (change of container, FFV1, FLAC), the checksum of every stream is the same.
  • The embedded metadata is carried over.
  • The delivered file stays as an original when the target is lossy.

Sources