Audio formats
Supported source audio formats for upload, and the output format values you can request in job targets.
Input formats
Create an asset with POST /assets, then upload the source bytes to the returned upload URL. The API accepts these source audio formats. Other formats are rejected when you create a job, with 415 UNSUPPORTED_FORMAT.
| Format | Extensions | Guidance |
|---|---|---|
WAV | .wav | Best default when you can choose the source format. Use 16- or 24-bit WAV up to 192 kHz. |
AIFF | .aiff, .aif | Use directly when AIFF is your source master; transcoding to WAV is not required. |
FLAC | .flac | Lossless compressed audio. Good when you need smaller uploads without lossy compression. |
MP3 | .mp3 | Accepted, but lossy compression artifacts can affect separation quality. |
AAC / M4A | .aac, .m4a | Accepted, but use WAV, AIFF, or FLAC when you can control the source format. |
Use the original source when possible
Transcoding a low-bitrate compressed file to WAV will not restore lost detail. When separation quality matters, upload the highest-quality original you have.
Sample rate and audio properties
The API accepts common production sample rates. Choose the source's native settings when you can; avoid upsampling or changing bit depth just to make a file look higher resolution.
| Property | Recommendation | Notes |
|---|---|---|
Sample rate | 44.1 kHz or 48 kHz is a good default; up to 192 kHz is accepted. | Use the source's native rate when possible. Upsampling a low-rate file does not add useful detail. |
Bit depth | 16-bit or 24-bit PCM for WAV and AIFF sources. | 24-bit is useful for production masters; 16-bit is fine for many delivery and review workflows. |
Channels | Mono and stereo are supported for every model; 5.1 is supported for DME models. | See 5.1 multichannel input below for the exact requirements. Any other channel count — and 5.1 sent to a model that does not accept it — is rejected with 422 UNSUPPORTED_CHANNELS. |
Maximum file size | 6 GB per upload. | Larger files are rejected with 413 FILE_TOO_LARGE. At 16-bit/48 kHz stereo PCM that is roughly 9 hours; a 24-bit/48 kHz 5.1 master fits about 2 hours, and a compressed source (FLAC, MP3) fits a much longer duration in the same 6 GB. |
5.1 multichannel input
DME separation accepts 5.1 (six-channel) input in addition to mono and stereo, and returns each separated track as a 5.1 file in standard SMPTE channel order. Every model listed under Dialogue, Music & Effects accepts 5.1, on every tier that model supports. Models in the other families accept mono and stereo only; GET /models reports the accepted layouts per model in input_channel_layouts.
| Requirement | Details |
|---|---|
Container | PCM WAV, RF64, or Wave64. A lossy six-channel file (for example AAC tagged as 5.1) is rejected — its channel labels cannot be trusted for a split. |
Channel layout | The file must carry an explicit 5.1 layout (5.1 or 5.1(side)). A file that only reports "6 channels" with no layout is rejected, because the channel order is then unknown. |
Targets | Every target in the job must accept 5.1 — check input_channel_layouts on the model in GET /models. You can request several such targets in one job; adding a stem, karaoke, or denoise target makes the whole job invalid. A tier can also narrow this: the Live tier of dme_dialogue_remove_vocals_v1 takes mono and stereo only, so naming it invalidates a 5.1 job even though the model itself accepts 5.1. |
A 5.1 job that fails any of these is rejected at POST /jobs with 422 UNSUPPORTED_CHANNELS, and the message names the condition that failed.
Not 5.1, and more than two channels?
Other multichannel layouts (for example 7.1 or a four-channel stem bundle) are not supported. Export the mix or the channel pair you want to separate before upload, or downmix to 5.1 when the source allows it.
Output formats
Each job target has a formats array. Use lowercase values, and request only the formats listed for the selected model in Models or GET /models. The current public model catalog exposes wav output.
| Value | Guidance |
|---|---|
wav | Available for all models in the current public catalog. Use for editing, review, and downstream audio processing. |
Selection guidance
For best results, avoid format conversion unless it solves a practical upload or compatibility problem. A lossless source such as WAV, AIFF, or FLAC is preferable when you can control the input.