Skip to content

vllm_omni.diffusion.models.soulx_singer.preprocess.metadata_utils

Metadata merge helpers for SoulX-Singer preprocess.

SegmentMetadata dataclass

end_time_ms instance-attribute 

end_time_ms: int

item_name instance-attribute 

item_name: str

language instance-attribute 

language: str

note_dur instance-attribute 

note_dur: list[float]

note_pitch instance-attribute 

note_pitch: list[int]

note_text instance-attribute 

note_text: list[str]

note_type instance-attribute 

note_type: list[int]

origin_wav_fn class-attribute instance-attribute 

origin_wav_fn: str | None = None

start_time_ms instance-attribute 

start_time_ms: int

wav class-attribute instance-attribute 

wav: ndarray | None = None

wav_fn instance-attribute 

wav_fn: str

get 

get(name: str, default: Any = None) -> Any

convert_metadata 

convert_metadata(
    item: SegmentMetadata, f0: ndarray | None = None
) -> dict[str, Any]

Convert merged segment metadata into SoulX JSON metadata format.

Matches upstream convert_metadata: F0 always comes from *_f0.npy written by extract_f0_file on the merged segment wav, not from an in-memory array.