data_juicer.ops.mapper.video_active_speaker_detect_mapper module#

class data_juicer.ops.mapper.video_active_speaker_detect_mapper.VideoActiveSpeakerDetectMapper(*args, **kwargs)[source]#

Bases: Mapper

Detect active speakers in a video by analyzing visual face tracks and audio signals, including consistency checks for gender and age.

Source: This operator is a part of HumanVBench (CVPR 2026).

__init__(temp_save_path: str = './temp_path', Light_ASD_model_path: str = './thirdparty/humanvbench_models/Light-ASD/weight/finetuning_TalkSet.model', active_threshold: int = 15, active_speaker_flag: str = 'active_speaker_flag', *args, **kwargs)[source]#

Initialization method.

Parameters:

blur_type

active_speaker_detection_revise(active_score, is_child_descrip, speech_audio, face_gender)[source]#
process_single(sample, rank=None)[source]#

For sample level, sample –> sample

Parameters:

sample – sample to process

Returns:

processed sample