Upload new model: vn-voice-l40s

.gitattributes

@@ -47,3 +47,4 @@ ellie-v3/generation_11.wav filter=lfs diff=lfs merge=lfs -text
 ember-v2/generation_4.wav filter=lfs diff=lfs merge=lfs -text
 ember-v2/generation_9.wav filter=lfs diff=lfs merge=lfs -text
 vn-voice-h200/4.wav filter=lfs diff=lfs merge=lfs -text
+vn-voice-l40s/4.wav filter=lfs diff=lfs merge=lfs -text

vn-voice-l40s/4.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6271035391e2193ff0760a5e340ebf46c255541585da2c10420672f5ad337757
+size 927356

vn-voice-l40s/config.yaml

@@ -0,0 +1,80 @@
+log_dir: ./Models/Finetune/vi—voice
+save_freq: 5
+log_interval: 10
+device: cuda
+epochs: 50 
+batch_size: 4
+max_len: 310 # maximum number of frames
+pretrained_model: ./Models/Finetune/base_model_120k_vi.pth
+load_only_params: true # set to true if do not want to load epoch numbers and optimizer parameters
+debug: true
+
+data_params:
+  train_data: "./Data/train_list.txt" 
+  val_data: "./Data/val_list.txt"
+  root_path: "./Data/24khz/"
+
+symbol: #Total 189 symbols
+  pad: "$"
+  punctuation: ';:,.!?¡¿—…"«»“” '
+  letters: "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
+  letters_ipa: "ɑɐɒæɓʙβɔɕçɗɖðʤəɘɚɛɜɝɞɟʄɡɠɢʛɦɧħɥʜɨɪʝɭɬɫɮʟɱɯɰŋɳɲɴøɵɸθœɶʘɹɺɾɻʀʁɽʂʃʈʧʉʊʋⱱʌɣɤʍχʎʏʑʐʒʔʡʕʢǀǁǂǃˈˌːˑʼʴʰʱʲʷˠˤ˞↓↑→↗↘'̩'ᵻ"
+  extend: "∫̆ăη͡123456" #ADD MORE SYMBOLS HERE
+
+preprocess_params:
+  sr: 24000
+  spect_params:
+    n_fft: 2048
+    win_length: 1200
+    hop_length: 300
+
+training_strats:
+  #All modules: 'decoder', 'predictor', 'text_encoder', 'style_encoder', 'text_aligner', 'pitch_extractor', 'mpd', 'msd'
+  freeze_modules: [''] # Not updated when training.
+  ignore_modules: [''] # Not loading => fresh start. IMPORTANT: 'text_aligner' and 'pitch_extractor' are util pretraineds DO NOT ignore them.
+
+model_params:
+  dim_in: 64 
+  hidden_dim: 512
+  max_conv_dim: 512
+  n_layer: 3
+  n_mels: 80
+  max_dur: 50 # maximum duration of a single phoneme
+  style_dim: 128 # style vector size
+  
+  dropout: 0.2
+
+  ASR_params:
+    input_dim: 80
+    hidden_dim: 256
+    n_layers: 6
+    token_embedding_dim: 512
+
+  JDC_params:
+    num_class: 1
+    seq_len: 192
+
+  # config for decoder
+  decoder: 
+      type: 'hifigan' # either hifigan or istftnet
+      resblock_kernel_sizes: [3,7,11]
+      upsample_rates :  [10,5,3,2]
+      upsample_initial_channel: 512
+      resblock_dilation_sizes: [[1,3,5], [1,3,5], [1,3,5]]
+      upsample_kernel_sizes: [20,10,6,4]
+  
+loss_params:
+    lambda_mel: 5. # mel reconstruction loss
+    lambda_gen: 1. # generator loss
+    
+    lambda_mono: 1. # monotonic alignment loss (TMA)
+    lambda_s2s: 1. # sequence—to—sequence loss (TMA)
+
+    lambda_F0: 1. # F0 reconstruction loss
+    lambda_norm: 1. # norm reconstruction loss
+    lambda_dur: 1. # duration loss
+    lambda_ce: 20. # duration predictor probability output CE loss
+
+optimizer_params:
+  lr: 0.0001 # general learning rate
+  ft_lr: 0.00001 # learning rate for acoustic modules

vn-voice-l40s/epoch_00049.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85073a78d2b2cecb5bdf28c89203568947cf87f2b5939abf7364e31f9b8251e3
+size 1692227939