Medix-AI/coqui-tts
Medix-AI
/
coqui-tts
mirror of https://github.com/coqui-ai/TTS.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

model config update

This commit is contained in:
erogol 2020-06-05 13:28:16 +02:00
parent 20a9d51ccc
commit 81af469be0
1 changed files with 14 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
vocoder/configs/multiband-melgan_and_rwd_config.json
View File

@ -49,21 +49,28 @@
// LOSS PARAMETERS // LOSS PARAMETERS
"use_stft_loss": true, "use_stft_loss": true,
"use_subband_stft_loss": true,
"use_mse_gan_loss": true, "use_mse_gan_loss": true,
"use_hinge_gan_loss": false, "use_hinge_gan_loss": false,
"use_feat_match_loss": false, // use only with melgan discriminators "use_feat_match_loss": false, // use only with melgan discriminators
"stft_loss_weight": 1, "stft_loss_weight": 0.5,
"subband_stft_loss_weight": 0.5,
"mse_gan_loss_weight": 2.5, "mse_gan_loss_weight": 2.5,
"hinge_gan_loss_weight": 1, "hinge_gan_loss_weight": 2.5,
"feat_match_loss_weight": 10.0, "feat_match_loss_weight": 25.0,
"stft_loss_params": { "stft_loss_params": {
"n_ffts": [1024, 2048, 512], "n_ffts": [1024, 2048, 512],
"hop_lengths": [120, 240, 50], "hop_lengths": [120, 240, 50],
"win_lengths": [600, 1200, 240] "win_lengths": [600, 1200, 240]
}, },
"target_loss": "avg_G_loss", // loss value to pick the best model "subband_stft_loss_params":{
"n_ffts": [384, 683, 171],
"hop_lengths": [30, 60, 10],
"win_lengths": [150, 300, 60]
},
"target_loss": "avg_G_loss",
// DISCRIMINATOR // DISCRIMINATOR
// "discriminator_model": "melgan_multiscale_discriminator", // "discriminator_model": "melgan_multiscale_discriminator",
@ -85,7 +92,7 @@
// GENERATOR // GENERATOR
"generator_model": "multiband_melgan_generator", "generator_model": "multiband_melgan_generator",
"generator_model_params": { "generator_model_params": {
"upsample_factors":[2 ,2, 4, 4], "upsample_factors":[8, 4, 2],
"num_res_blocks": 4 "num_res_blocks": 4
}, },
@ -111,7 +118,7 @@
"noam_schedule": false, // use noam warmup and lr schedule. "noam_schedule": false, // use noam warmup and lr schedule.
"warmup_steps_gen": 4000, // Noam decay steps to increase the learning rate from 0 to "lr" "warmup_steps_gen": 4000, // Noam decay steps to increase the learning rate from 0 to "lr"
"warmup_steps_disc": 4000, "warmup_steps_disc": 4000,
"epochs": 1000, // total number of epochs to train. "epochs": 100000, // total number of epochs to train.
"wd": 0.000001, // Weight decay weight. "wd": 0.000001, // Weight decay weight.
"lr_gen": 0.0001, // Initial learning rate. If Noam decay is active, maximum learning rate. "lr_gen": 0.0001, // Initial learning rate. If Noam decay is active, maximum learning rate.
"lr_disc": 0.0001, "lr_disc": 0.0001,
@ -131,6 +138,6 @@
"eval_split_size": 10, "eval_split_size": 10,
// PATHS // PATHS
"output_path": "/home/erogol/Models/LJSpeech/" "output_path": "/data/rw/home/Trainings/"
} }