Skip to content

SOTA Open Source TTS

Update larger config

Lengyue65f811ba4115
diff --git a/fish_speech/configs/vqgan_pretrain_v2.yaml b/fish_speech/configs/vqgan_pretrain_v2.yaml
index 8691020..385d814 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_large
+project: vqgan_pretrain_v2_large_30
# Lightning Trainer
trainer:
@@ -64,7 +64,7 @@ model:
encoder:
_target_: fish_speech.models.vqgan.modules.modules.WaveNet
- hidden_channels: 384
+ hidden_channels: 512
kernel_size: 3
dilation_rate: 2
n_layers: 20
@@ -72,16 +72,16 @@ model:
vq:
_target_: fish_speech.models.vqgan.modules.encoders.VQEncoder
- in_channels: 384
- vq_channels: 384
+ in_channels: 512
+ vq_channels: 512
codebook_size: 256
- codebook_groups: 2
+ codebook_groups: 4
codebook_layers: 2
downsample: 4
decoder:
_target_: fish_speech.models.vqgan.modules.modules.WaveNet
- hidden_channels: 384
+ hidden_channels: 512
kernel_size: 3
dilation_rate: 2
n_layers: 20