Nodes/ComfyUI-FLOAT_Optimized/FLOAT Audio Feature Extract (VA)
ComfyUI Node

FLOAT Audio Feature Extract (VA)

Processes a batch of pre-validated (mono, correct SR) audio. It applies the feature extractor from the loaded Wav2Vec pipe, runs the audio through the Wav2Vec model, and interpolates the resulting features to match the target video FPS, making them ready for projection.

By set-soft·Created about a year ago·Updated 12 months ago· 33
FLOAT Audio Feature Extract (VA)
  • audio
  • wav2vec_pipe
  • wav2vec_features
  • audio_num_frames
  • processed_audio_features
  • wav2vec_pipe_out
  • audio
  • fps
target_fps25.0
only_last_featuresfalse
CategoryFLOAT/Very Advanced

Inputs (4)

NameTypeDefaultDescription
audioAUDIOThe raw ComfyUI audio input. Must be mono and have the correct sample rate required by the Wav2Vec pipe.
wav2vec_pipeWAV2VEC_PIPEThe loaded Wav2Vec pipe, containing the model, feature extractor, and options.
target_fpsFLOAT25.0The target video frames-per-second. Used to calculate the final number of feature frames.
only_last_featuresBOOLEANfalseIf True, use only the features from the last transformer layer. If False, concatenate features from all transformer layers, resulting in a much larger feature dimension.

Outputs (6)

NameTypeDescription
wav2vec_featuresTORCH_TENSOR
audio_num_framesINT
processed_audio_featuresTORCH_TENSOR
wav2vec_pipe_outWAV2VEC_PIPE
audioAUDIO
fpsFLOAT