Skip to content

SOTA Open Source TTS

Add larger config

Lengyueeec6cbb2e080
diff --git a/fish_speech/configs/vqgan_pretrain_v2.yaml b/fish_speech/configs/vqgan_pretrain_v2.yaml
index ea3fad6..8691020 100644
--- a/fish_speech/configs/vqgan_pretrain_v2.yaml
+++ b/fish_speech/configs/vqgan_pretrain_v2.yaml
@@ -2,7 +2,7 @@ defaults:
- base
- _self_
-project: vqgan_pretrain_v2
+project: vqgan_pretrain_v2_large
# Lightning Trainer
trainer:
@@ -67,7 +67,7 @@ model:
hidden_channels: 384
kernel_size: 3
dilation_rate: 2
- n_layers: 10
+ n_layers: 20
in_channels: ${num_mels}
vq:
@@ -84,7 +84,7 @@ model:
hidden_channels: 384
kernel_size: 3
dilation_rate: 2
- n_layers: 10
+ n_layers: 20
out_channels: ${num_mels}
generator:
@@ -99,7 +99,7 @@ model:
use_template: true
pre_conv_kernel_size: 7
post_conv_kernel_size: 7
- ckpt_path: checkpoints/hifigan-base-comb-mix-lb-020/step_001200000_weights_only.ckpt
+ ckpt_path: checkpoints/hifi-gan-base-002000000.ckpt
discriminator:
_target_: fish_speech.models.vqgan.modules.modules.WaveNet