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

fixinf discriminator configuration

This commit is contained in:
erogol 2020-06-17 11:13:16 +02:00
parent 55230bb317
commit df10e2a92a
2 changed files with 3 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vocoder/configs/multiband_melgan_config.json
View File

@ -78,8 +78,8 @@
"discriminator_model": "melgan_multiscale_discriminator", "discriminator_model": "melgan_multiscale_discriminator",
"discriminator_model_params":{ "discriminator_model_params":{
"base_channels": 16, "base_channels": 16,
"max_channels":1024, "max_channels":512,
"downsample_factors":[4, 4, 4, 4] "downsample_factors":[4, 4, 4]
}, },
"steps_to_start_discriminator": 200000, // steps required to start GAN trainining.1 "steps_to_start_discriminator": 200000, // steps required to start GAN trainining.1
@ -110,9 +110,6 @@
"test_sentences_file": null, // set a file to load sentences to be used for testing. If it is null then we use default english sentences. "test_sentences_file": null, // set a file to load sentences to be used for testing. If it is null then we use default english sentences.
// OPTIMIZER // OPTIMIZER
"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_disc": 4000, // Noam decay steps to increase the learning rate from 0 to "lr"
"epochs": 10000, // total number of epochs to train. "epochs": 10000, // total number of epochs to train.
"wd": 0.0, // Weight decay weight. "wd": 0.0, // Weight decay weight.
"gen_clip_grad": -1, // Generator gradient clipping threshold. Apply gradient clipping if > 0 "gen_clip_grad": -1, // Generator gradient clipping threshold. Apply gradient clipping if > 0

2
vocoder/models/melgan_multiscale_discriminator.py
View File

@ -11,7 +11,7 @@ class MelganMultiscaleDiscriminator(nn.Module):
kernel_sizes=(5, 3), kernel_sizes=(5, 3),
base_channels=16, base_channels=16,
max_channels=1024, max_channels=1024,
downsample_factors=(4, 4, 4, 4), downsample_factors=(4, 4, 4),
pooling_kernel_size=4, pooling_kernel_size=4,
pooling_stride=2, pooling_stride=2,
pooling_padding=1): pooling_padding=1):