Audio glitches

[CMoranoA]

Hi! First of all, thank you for your kind words about this library, much appreciated! I've encountered a similar issue with certain audio files. After applying deep filtering, I've noticed outlier short noises, almost like glitches. Has anyone else experienced this? Thanks!

[shenxiaozheng]

我也遇到这样的问题，特别是噪声刚刚出现的时刻。

[viki347]

I'm seeing a similar issue with DeepFilterNet2. How do I need to modify it? I would greatly appreciate it if you could help me！ Here is my configuration file： [train] seed = 42 device = model = deepfilternet2 mask_only = False df_only = False jit = False batch_size = 96 batch_size_eval = 0 overfit = False num_workers = 4 max_sample_len_s = 5.0 num_prefetch_batches = 32 global_ds_sampling_f = 1.0 dataloader_snrs = -5,0,5,10,20,40 dataloader_gains = -6,0,6 batch_size_scheduling = 0/8, 1/16, 2/24, 4/32, 9/64, 19/96 max_epochs = 50 validation_criteria = loss validation_criteria_rule = min early_stopping_patience = 5 start_eval = False

[df] sr = 48000 fft_size = 960 hop_size = 480 nb_erb = 32 nb_df = 96 norm_tau = 1 lsnr_max = 35 lsnr_min = -15 min_nb_erb_freqs = 2 df_order = 5 df_lookahead = 0 pad_mode = input

[deepfilternet] conv_lookahead = 0 conv_ch = 16 conv_depthwise = True convt_depthwise = True conv_kernel = 1,3 conv_kernel_inp = 3,3 emb_hidden_dim = 256 emb_num_layers = 2 df_hidden_dim = 256 df_gru_skip = none df_output_layer = linear df_pathway_kernel_size_t = 1 enc_concat = False df_num_layers = 3 df_n_iter = 2 gru_type = grouped gru_groups = 1 linear_groups = 1 group_shuffle = True dfop_method = real_unfold mask_pf = False

[distortion] p_reverb = 0.5 p_bandwidth_ext = 0.0 p_clipping = 0.0 p_zeroing = 0.0 p_air_absorption = 0.5 p_interfer_sp = 0.0

[optim] lr = 0.0005 momentum = 0 weight_decay = 0.05 optimizer = adamw lr_min = 1e-06 lr_warmup = 0.0001 warmup_epochs = 3 lr_cycle_mul = 1.0 lr_cycle_decay = 0.5 lr_cycle_epochs = -1 weight_decay_end = -1

[maskloss] factor = 0 mask = iam gamma = 0.6 gamma_pred = 0.6 f_under = 2 max_freq = 0

[spectralloss] factor_magnitude = 0 factor_complex = 0 factor_under = 1 gamma = 1

[multiresspecloss] factor = 0 factor_complex = 0 gamma = 1 fft_sizes = 512,1024,2048

[sdrloss] factor = 0

[localsnrloss] factor = 0.0005

[asrloss] factor = 0 factor_lm = 0 loss_lm = CrossEntropy model = base.en

[viki347]

I just solved this problem. I'll share it here. Through inspection, I found that my problem was due to incorrect configuration of loss. I estimate that the loss parameters in the config file generated by the author are default and they need to be modified according to the papers. Here are some reference links： https://github.com/Rikorose/DeepFilterNet/issues/130

[CMoranoA]

Hi Viki, can you please expand a little bit your solution? Thank you!

[github-actions[bot]]

This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 7 days.