Skip to main content
LALAL.AI is a dedicated stem separation service built on a set of proprietary neural networks, capable of splitting a track into a dozen distinct stem types. In an Ollang dubbing pipeline it does one specific and important job: separate the source audio into an isolated dialogue stem and a music-and-effects (M&E) bed. The dialogue stem feeds transcription; the M&E bed is preserved and remixed under the synthesized target-language voice. That separation is what makes a dub sound like a dub rather than a voiceover talking over the original mix.
Available in Ollang Workflows as lalalAi — AI Dubbing · Default. See the Audio Processing catalog.

​
Separation networks

The v1 API exposes six named splitters plus an auto option that selects the most effective one for the requested stem: phoenix · orion · perseus · andromeda · lyra · lynx · auto Orion is the one that matters most for dubbing quality. Rather than carving stems out of the mix with spectral masks, it synthesizes them — reconstructing the stem from what it has learned across many mixes. The practical result is markedly less of the artifact residue that betrays a dub: the watery, phase-smeared remainder left where dialogue used to be. Unless you have a reason to pin a specific network, auto is the sensible default: the best splitter for a stem is not the same across all of them.

​
Extractable stems

The API accepts twelve stem values: vocals · voice · music · drum · bass · piano · guitar · electric_guitar · acoustic_guitar · synthesizer · strings · wind vocals and voice are separate values — the first targets sung vocals, the second spoken voice, which is the one a dubbing pipeline wants. music is its counterpart: the bed that has to survive under the new dialogue. guitar is a generic value alongside the more specific electric_guitar and acoustic_guitar. For dubbing, the relevant split is voice against music. The finer stems matter when a mix has to be partially rebuilt — for example, when original dialogue sits under a music cue that must be preserved intact.

​
Processing options

​
Where it fits in a workflow

LALAL.AI runs on every AI Dubbing Order and is the default for the split step. It is the shortest stage in the pipeline, and the one with the largest effect on whether the finished mix sounds professional. Dereverberation is worth knowing about specifically: reverb in the source is one of the more reliable ways to degrade transcription accuracy downstream, so cleaning it at the separation step improves every subsequent stage.

​
Considerations

  • Separation quality tracks source quality. Heavily compressed audio, dialogue buried under a loud mix, or material already processed by another separation tool will produce a weaker M&E bed.
  • A clean stem always beats a separated one. Where the source has a discrete dialogue stem available, supplying it directly gives a better result than separating it back out.

​
Reference