mrq/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

I don't know if the lack of start/stop tokens being added was causing my inference tests to fail, but it seems better now

Browse Source
This commit is contained in:
mrq 2023-08-20 19:21:54 -05:00
parent 736c077282
commit a47029065b
2 changed files with 12 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vall_e/data.py
View File

@ -28,8 +28,8 @@ from tqdm.auto import tqdm
_logger = logging.getLogger(__name__)
def get_phone_symmap():
if cfg.dataset.use_hdf5 and 'symmap' in cfg.hdf5:
return json.loads( cfg.hdf5['symmap'].asstr()[()] )
#if cfg.dataset.use_hdf5 and 'symmap' in cfg.hdf5:
# return json.loads( cfg.hdf5['symmap'].asstr()[()] )
symmap = {'<s>': 1, '</s>': 2, ' ': 3, '.': 4, ',': 5, '!': 6, '?': 7, 'p': 7, 'iː': 8, 'ɚ': 9, 'ˌ': 10, 'dˌ': 11, 'mˌ': 12, 'd': 13, 'ɹ': 14, 'tˈ': 15, 'pˌ': 16, 'uː': 17, 'l': 18, 'æ': 19, 'ɛ': 20, 'ɪ': 21, 'j': 22, 'ʊ': 23, 't': 24, 'n': 25, 'v': 26, 'a': 27, 'o': 28, 'ŋ': 29, 'w': 30, 'ʌ': 31, 'hˈ': 32, 'ɡˈ': 33, 'ə': 34, 'θˈ': 35, 'dˈ': 36, 'wˌ': 37, 'h': 38, 'z': 39, 'k': 40, 'ð': 41, 'ɡˌ': 42, 'ˈ': 43, 'fˈ': 44, 'i': 45, 's': 46, 'ʃ': 47, 'wˈ': 48, 'ðˈ': 49, 'ɹˈ': 50, 'lˈ': 51, 'ɡ': 52, 'oː': 53, 'mˈ': 54, 'e': 55, 'ɑː': 56, 'nˈ': 57, 'm': 58, 'θˌ': 59, 'sˈ': 60, 'f': 61, 'ɔː': 62, 'hˌ': 63, 'b': 64, 'jˈ': 65, 'ɐ': 66, 'ʒˈ': 67, 'θ': 68, 'bˈ': 69, 'ɾ': 70, 'ɜː': 71, 'ʌˈ': 72, 'ʃˌ': 73, 'bˌ': 74, 'kˈ': 75, 'ɔ': 76, 'zˈ': 77, 'ᵻ': 78, 'kˌ': 79, 'vˈ': 80, 'fˌ': 81, 'ʒ': 82, 'ʃˈ': 83, 'ɹˌ': 84, 'tˌ': 85, 'pˈ': 86, 'ðˌ': 87, 'sˌ': 88, 'nˌ': 89, 'lˌ': 90, '̩': 91, 'ʔ': 92, 'vˌ': 93, 'ɪˈ': 94, '"': 95, 'ɪˌ': 96, 'ʒˌ': 97, 'uːˌ': 98, 'ʊˈ': 99, 'jˌ': 100, 'uːˈ': 101, 'iːˈ': 102, 'zˌ': 103, '.ˈ': 104, '…': 105, 'ŋˌ': 106, 'ɐˌ': 107, '—ˈ': 108, 'iˌ': 109, 'iːˌ': 110, 'ɛː': 111, ')': 112, ')ˈ': 113, '(': 114, 'u': 115, '-': 116, 'ɖˈ': 117, 'iˈ': 118, 'ʰˈ': 119, 'ɟˈ': 120, '̃': 121, 'eː': 122, 'ɾˈ': 123, 'r': 124, 'ʰ': 125, '-ˌ': 126, 'ɫ': 127, 'q': 128, '—': 129, 'ʊˌ': 130, 'aː': 131, 'cˈ': 132, '…ˈ': 133, 'c': 134, 'ɳ': 135, 'ɐˈ': 136, 'x': 137, 'ʔˌ': 138, '.ˌ': 139, 'ɑ': 140, '?ˈ': 141, '̩ˈ': 142, '"ˈ': 143, ',ˈ': 144, 'ŋˈ': 145, 'əˌ': 146, '!ˈ': 147, '"ˌ': 148, '?ˌ': 149, ',ˌ': 150, '—ˌ': 151, '̩ˌ': 152, 'əˈ': 153, '!ˌ': 154, 'ɬ': 155, 'ʲ': 156, '¡': 157, 'ɯ': 158, 'qˌ': 159, 'ʑ': 160, 'ʑˈ': 161, '¿': 162, 'ɑːˈ': 163, 'iːː': 164, 'ɛˈ': 165, '¡ˈ': 166, 'æˈ': 167, 'ç': 168, 'ɾˌ': 169, 'ᵻˈ': 170, 'xˈ': 171, 'ɔːˈ': 172, ';': 173, 'ɬˌ': 174, ':': 175, 'ʔˈ': 176, 'ɑːˌ': 177, 'ɬˈ': 178}
return symmap

17
vall_e/inference.py
View File

@ -24,7 +24,10 @@ class TTS():
if config:
cfg.load_yaml( config )
cfg.format()
try:
cfg.format()
except Exception as e:
pass
"""
if cfg.trainer.load_state_dict:
@ -72,12 +75,12 @@ class TTS():
self.nar = engine.module.to(self.device)
def encode_text( self, text, lang_marker="en" ):
text = g2p.encode(text)
phones = [f"<{lang_marker}>"] + [ " " if not p else p for p in text ] + [f"</{lang_marker}>"]
mapped = [self.symmap[p] for p in phones if p in self.symmap]
return torch.tensor( mapped )
content = g2p.encode(text)
#phones = ["<s>"] + [ " " if not p else p for p in content ] + ["</s>"]
phones = [ " " if not p else p for p in content ]
return torch.tensor([ 1 ] + [*map(self.symmap.get, phones)] + [ 2 ])
def encode_audio( self, path ):
def encode_audio( self, path, trim=True ):
enc = qnt.encode_from_file( path )
res = enc[0].t().to(torch.int16)
if trim:
@ -85,7 +88,7 @@ class TTS():
return res
def inference( self, text, reference, mode="both", max_ar_steps=6 * 75, ar_temp=1.0, nar_temp=1.0, out_path="./.tmp.wav" ):
def inference( self, text, reference, max_ar_steps=6 * 75, ar_temp=1.0, nar_temp=1.0, out_path="./.tmp.wav" ):
prom = self.encode_audio( reference )
phns = self.encode_text( text )