core.models.audio.packed_audio

Module Contents

Classes

PackedAudioEmbeddings

Flat valid audio embeddings with one length per source audio.

API

class core.models.audio.packed_audio.PackedAudioEmbeddings

Flat valid audio embeddings with one length per source audio.

embeddings: torch.Tensor

None

lengths: torch.Tensor

None

property cu_seqlens: torch.Tensor

Return cumulative sequence lengths with a leading zero (int32).

pad_to_lengths(

target_lengths: torch.Tensor,

) → core.models.audio.packed_audio.PackedAudioEmbeddings

Zero-pad each packed embedding up to the given per-source target lengths.