#!/bin/bash mode: 'inference' use_cuda: 1 # 1 for True, 0 for False num_gpu: 1 sampling_rate: 16000 network: "MossFormer2_SS_16K" ##network type checkpoint_dir: "checkpoints/MossFormer2_SS_16K" input_path: "data/wsj0_2mix_16k_fullpath.lst" # an input dir or input scp file #input_path: "/home/shengkui.zhao/DingTalk_NS/data/webrtc_test_set_20200521_16k/noisy" #input_path: "/mnt/nas_sg/mit_sg/shengkui.zhao/ComplexNN/audio/youtube_testset_16k/noisy_long/noisy" output_dir: "outputs/MossFormer2_SS_16K_wsj0_2mix" ## output dir to store processed audio # decode parameters one_time_decode_length: 30 #maximum segment length for one-pass decoding (seconds), longer audio (>3s) will use segmented decoding decode_window: 10 #one-pass decoding length num_spks: 2 encoder_kernel_size: 16 encoder_embedding_dim: 512 mossformer_sequence_dim: 512 num_mossformer_layer: 24