| # @package experiment | |
| name: seanet-nfft=${experiment.nfft}-hl=${experiment.hop_length} | |
| # Dataset related | |
| lr_sr: 44100 # low resolution sample rate, added to support BWE. Should be included in training cfg | |
| hr_sr: 44100 # high resolution sample rate. Should be included in training cfg | |
| segment: 2 | |
| stride: 2 # in seconds, how much to stride between training examples | |
| pad: true # if training sample is too short, pad it | |
| upsample: false | |
| batch_size: 5 | |
| nfft: 1024 | |
| hop_length: 512 | |
| # models related | |
| model: seanet | |
| seanet: | |
| latent_space_size: 256 | |
| ngf: 32 | |
| n_residual_layers: 4 | |
| resample: 1 | |
| normalize: False | |
| floor: 0.0005 | |
| ratios: [ 8,8,2,2 ] | |
| lr_sr: 44100 | |
| hr_sr: 44100 | |
| in_channels: 2 | |
| out_channels: 2 | |
| adversarial: True | |
| features_loss_lambda: 100 | |
| discriminator_models: [ msd, mpd ] #msd_melgan/msd_hifi/mpd/hifi | |
| only_features_loss: False | |
| only_adversarial_loss: False | |
| msd: | |
| hidden: 16 | |
| channels: 2 | |
| mpd: | |
| hidden: 16 | |
| channels: 2 | |
| melgan_discriminator: | |
| n_layers: 5 | |
| num_D: 3 | |
| downsampling_factor: 1 | |
| ndf: 16 | |
| channels: 2 |