Media¶

How an audio file in the media pool is recorded, and what compose writes when it points a template's media-pool item at your file. Values are read with ffprobe; nothing is guessed — unsupported files are refused.

Supported formats¶

Container	`ffprobe` codec	Bit depth	`CodecName`	`BitDepth`	Clip field 18 (format flags)
WAV	`pcm_s16le`	16-bit	`Linear PCM`	1	32768 (`0x8000`)
WAV	`pcm_s24le`	24-bit	`Linear PCM`	2	32768 (`0x8000`)
WAV	`pcm_f32le`	32-bit float	`Linear PCM`	3	32768 (`0x8000`)
FLAC	`flac`	16-bit	`FLAC`	1	32778 (`0x800A`)
FLAC	`flac`	24-bit	`FLAC`	2	32778 (`0x800A`)
M4A	`aac` (profile LC)	—	`AAC`	3	32772 (`0x8004`)
MP3	`mp3`	—	`MP3`	3	32776 (`0x8008`)

BitDepth: 1 = 16-bit, 2 = 24-bit, 3 = 32-bit float. Compressed formats are stored as 3.
Sample rate 44100 or 48000 Hz; 1 or 2 channels; exactly one audio stream.
Anything else (other codecs, 8/32-bit integer PCM, other rates, more channels, video files) is rejected.

The media-pool item (`Sm2MpAudioClip`)¶

In MediaPool/**/MpFolder.xml:

Field	Encoding	Content
`Name`	text	file name
`UniqueMediaPoolItemId`	UUID	item id; a fresh one for each extra media file `compose` adds
`Clip`	versioned blob, protobuf	file location and format (below)
`TracksBA`	QDataStream	per-stream audio description (below)
`FieldsBlob`	versioned blob, protobuf	embedded audio info (below)
`EmbeddedAudioVec` → `BtAudioInfo`	XML	the embedded audio stream record (`DbId` referenced from the channel mapping)
`VirtualAudioTracksBA`	hex	virtual audio track layout

`Clip` (protobuf)¶

Field	Type	Content
1	text	folder of the file
2	text	file name
3	text	modification time, C `ctime` format (`Sun Oct 18 17:34:12 2026`)
5	text	codec name (`Linear PCM`, `FLAC`, `AAC`, `MP3`)
13	varint	modification time, microseconds since the epoch
15, 16	varint	not decoded (kept)
18	varint	format flags (table above)

`TracksBA` (QDataStream)¶

A QVariantMap with one key, "0", whose QByteArray value is a nested QVariantMap describing the first audio stream:

Key	Content
`UniqueId`	stream id
`DbType`	`BtAudioTrack`
`IdxTrack`	stream index (0)
`StartTime`	0.0
`SampleRate`	Hz
`NumChannels`	1 or 2
`Duration`	length in samples
`CodecName`	as in `Clip` field 5
`BitDepth`	1 / 2 / 3
`ChannelLayout`	channel layout; present when `ffprobe` reports `mono` / `stereo`

`FieldsBlob` (protobuf)¶

Field 1 → field 5 holds the embedded audio info, again as a one-entry QVariantMap around a nested map:

Key	Content
`MediaExtents`	two little-endian doubles: 0 and the duration in seconds
`CodecName`	codec name
`SampleRate`	Hz — omitted at 48000
`BitDepth`	1 / 2 — omitted for 3 (32-bit is the default in every capture)
`Type`	0 for stereo; omitted for mono
`ChannelVecBA`	channel mapping: a map `{"0": {MediaRef}, "1": {MediaRef, ChannelIdx: 1}}`, one entry per channel, `MediaRef` = the `BtAudioInfo` `DbId`

Durations¶

Duration (samples) is the stream's duration_ts × time_base × rate (or duration × rate).
MP3: Resolve counts all encoded MP3 frames, unlike ffprobe's gapless duration, so drpkit sums the packet durations instead.
source_out in the cut list may not exceed the shorter of the two durations (stored length and playable content).

Channel changes¶

When the new file has a different channel count than the template's media (mono ↔ stereo), compose also rewrites NumChannels, ChannelVecBA, Type and each clip's VirtualAudioTrackBA (ChannelsBA = u32BE 2, count + 0x4001 + i per channel; AudioType 1 for mono, 0 for stereo). It does so only when the template uses Resolve's default channel mapping; otherwise the change is refused.

Several files on one track¶

A cut row with its own media creates a new media-pool item, cloned from the track's template item under fresh ids (item DbId, embedded BtAudioInfo DbId — also renamed inside the channel mapping, where it is stored as UTF-16 — and UniqueMediaPoolItemId), then filled with that file's metadata. The same file on two tracks, and one media-pool item shared by two template tracks, are refused.

Verification¶

Media metadata for all 28 combinations (WAV 16/24/32f, FLAC 16/24, AAC, MP3 × 44.1/48 kHz × mono/stereo) is compared field by field with Resolve's own exports of the same files (tests/compose/formats/). 56 composed projects covering all 28 combinations were imported into Resolve 21.1.1 with every clip Online at the requested positions, In points and fades, audible on both channels.