pradnya-hf-dev
commited on
Commit
•
182ff19
1
Parent(s):
993b839
update with spn_predictor
Browse files- hyperparams.yaml +23 -4
hyperparams.yaml
CHANGED
@@ -2,6 +2,7 @@
|
|
2 |
# Model: Fastspeech2 for TTS
|
3 |
# Authors: Sathvik Udupa, Yingzhi Wang, Pradnya Kandarkar
|
4 |
# ################################
|
|
|
5 |
# Input parameters
|
6 |
lexicon:
|
7 |
- AA
|
@@ -45,7 +46,7 @@ lexicon:
|
|
45 |
- ZH
|
46 |
- spn
|
47 |
|
48 |
-
n_symbols: 41 #fixed
|
49 |
padding_idx: 0
|
50 |
n_mel_channels: 80
|
51 |
|
@@ -73,18 +74,34 @@ postnet_kernel_size: 5
|
|
73 |
postnet_n_convolutions: 5
|
74 |
postnet_dropout: 0.5
|
75 |
|
76 |
-
#
|
77 |
normalize_before: True
|
78 |
ffn_type: 1dcnn #1dcnn or ffn
|
79 |
ffn_cnn_kernel_size_list: [9, 1]
|
80 |
|
81 |
-
#
|
82 |
dur_pred_kernel_size: 3
|
83 |
pitch_pred_kernel_size: 3
|
84 |
energy_pred_kernel_size: 3
|
85 |
variance_predictor_dropout: 0.5
|
86 |
|
87 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
model: !new:speechbrain.lobes.models.FastSpeech2.FastSpeech2
|
89 |
enc_num_layers: !ref <enc_num_layers>
|
90 |
enc_num_head: !ref <enc_num_head>
|
@@ -119,8 +136,10 @@ model: !new:speechbrain.lobes.models.FastSpeech2.FastSpeech2
|
|
119 |
input_encoder: !new:speechbrain.dataio.encoder.TextEncoder
|
120 |
|
121 |
modules:
|
|
|
122 |
model: !ref <model>
|
123 |
|
124 |
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
125 |
loadables:
|
|
|
126 |
model: !ref <model>
|
|
|
2 |
# Model: Fastspeech2 for TTS
|
3 |
# Authors: Sathvik Udupa, Yingzhi Wang, Pradnya Kandarkar
|
4 |
# ################################
|
5 |
+
|
6 |
# Input parameters
|
7 |
lexicon:
|
8 |
- AA
|
|
|
46 |
- ZH
|
47 |
- spn
|
48 |
|
49 |
+
n_symbols: 41 #fixed depending on symbols in the lexicon +1 for a dummy symbol used for padding
|
50 |
padding_idx: 0
|
51 |
n_mel_channels: 80
|
52 |
|
|
|
74 |
postnet_n_convolutions: 5
|
75 |
postnet_dropout: 0.5
|
76 |
|
77 |
+
# common
|
78 |
normalize_before: True
|
79 |
ffn_type: 1dcnn #1dcnn or ffn
|
80 |
ffn_cnn_kernel_size_list: [9, 1]
|
81 |
|
82 |
+
# variance predictor
|
83 |
dur_pred_kernel_size: 3
|
84 |
pitch_pred_kernel_size: 3
|
85 |
energy_pred_kernel_size: 3
|
86 |
variance_predictor_dropout: 0.5
|
87 |
|
88 |
+
# SPN predictor
|
89 |
+
spn_predictor: !new:speechbrain.lobes.models.FastSpeech2.SPNPredictor
|
90 |
+
enc_num_layers: !ref <enc_num_layers>
|
91 |
+
enc_num_head: !ref <enc_num_head>
|
92 |
+
enc_d_model: !ref <enc_d_model>
|
93 |
+
enc_ffn_dim: !ref <enc_ffn_dim>
|
94 |
+
enc_k_dim: !ref <enc_k_dim>
|
95 |
+
enc_v_dim: !ref <enc_v_dim>
|
96 |
+
enc_dropout: !ref <enc_dropout>
|
97 |
+
normalize_before: !ref <normalize_before>
|
98 |
+
ffn_type: !ref <ffn_type>
|
99 |
+
ffn_cnn_kernel_size_list: !ref <ffn_cnn_kernel_size_list>
|
100 |
+
n_char: !ref <n_symbols>
|
101 |
+
padding_idx: !ref <padding_idx>
|
102 |
+
|
103 |
+
|
104 |
+
#model
|
105 |
model: !new:speechbrain.lobes.models.FastSpeech2.FastSpeech2
|
106 |
enc_num_layers: !ref <enc_num_layers>
|
107 |
enc_num_head: !ref <enc_num_head>
|
|
|
136 |
input_encoder: !new:speechbrain.dataio.encoder.TextEncoder
|
137 |
|
138 |
modules:
|
139 |
+
spn_predictor: !ref <spn_predictor>
|
140 |
model: !ref <model>
|
141 |
|
142 |
pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
|
143 |
loadables:
|
144 |
+
spn_predictor: !ref <spn_predictor>
|
145 |
model: !ref <model>
|