deviandice/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Move gpt-tts back to lucidrains implementation

Browse Source 
Much better performance.
This commit is contained in:
James Betker 2021-08-05 22:15:13 -06:00
parent d120e1aa99
commit 89d15c9e74
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
codes/models/gpt_voice/gpt_tts.py
View File

@ -28,8 +28,8 @@ class GptTts(nn.Module):
self.mel_embedding = nn.Embedding(self.MEL_DICTIONARY_SIZE, model_dim) self.mel_embedding = nn.Embedding(self.MEL_DICTIONARY_SIZE, model_dim)
self.text_pos_embedding = nn.Embedding(self.MAX_SYMBOLS_PER_PHRASE, model_dim) self.text_pos_embedding = nn.Embedding(self.MAX_SYMBOLS_PER_PHRASE, model_dim)
self.mel_pos_embedding = nn.Embedding(max_mel_frames, model_dim) self.mel_pos_embedding = nn.Embedding(max_mel_frames, model_dim)
self.gpt = GPT(GPTConfig(1+self.MAX_SYMBOLS_PER_PHRASE+max_mel_frames, n_layer=8, n_embd=model_dim, n_head=8), do_pos_emb=False) #self.gpt = GPT(GPTConfig(1+self.MAX_SYMBOLS_PER_PHRASE+max_mel_frames, n_layer=8, n_embd=model_dim, n_head=8), do_pos_emb=False)
#self.gpt = Transformer(dim=model_dim, depth=8, seq_len=1+self.MAX_SYMBOLS_PER_PHRASE+max_mel_frames, heads=8) self.gpt = Transformer(dim=model_dim, depth=8, seq_len=1+self.MAX_SYMBOLS_PER_PHRASE+max_mel_frames, heads=8)
self.final_norm = nn.LayerNorm(model_dim) self.final_norm = nn.LayerNorm(model_dim)
self.text_head = nn.Linear(model_dim, self.NUMBER_TEXT_TOKENS) self.text_head = nn.Linear(model_dim, self.NUMBER_TEXT_TOKENS)