diff --git a/vall_e/config.py b/vall_e/config.py
index 16febda..f81bccc 100755
--- a/vall_e/config.py
+++ b/vall_e/config.py
@@ -159,7 +159,7 @@ class Model:
size: str = "full"
resp_levels: int = 1
prom_levels: int = 8
- tasks: int = 1 # 8 # ["tts", "ns", "sr", "tse", "cse", "nse"] and leaves two more for anything else I want (like "svc")
+ tasks: int = 8 # ["tts", "ns", "sr", "tse", "cse", "nse"] and leaves two more for anything else I want (like "svc")
arch_type: str = "transformer"
@property
@@ -217,8 +217,8 @@ class Models:
_max_levels: int = 0
_models: list[Model] = field(default_factory=lambda: [
- Model(name="ar", resp_levels=1, prom_levels=8, tasks=1),
- Model(name="nar", resp_levels=7, prom_levels=8, tasks=1),
+ Model(name="ar", resp_levels=1, prom_levels=8, tasks=8),
+ Model(name="nar", resp_levels=7, prom_levels=8, tasks=8),
])
def get(self, name=None):
@@ -503,6 +503,10 @@ class Config(_Config):
self.bitsandbytes = BitsAndBytes(**self.bitsandbytes)
self.trainer.deepspeed = DeepSpeed(**self.trainer.deepspeed)
+
+ self.dataset.training = [ Path(dir) for dir in self.dataset.training ]
+ self.dataset.validation = [ Path(dir) for dir in self.dataset.validation ]
+ self.dataset.noise = [ Path(dir) for dir in self.dataset.noise ]
cfg = Config.from_cli()
@@ -515,9 +519,6 @@ try:
if cfg.dataset.use_hdf5:
cfg.load_hdf5()
- cfg.dataset.training = [ Path(dir) for dir in cfg.dataset.training ]
- cfg.dataset.validation = [ Path(dir) for dir in cfg.dataset.validation ]
- cfg.dataset.noise = [ Path(dir) for dir in cfg.dataset.noise ]
except Exception as e:
pass
diff --git a/vall_e/data.py b/vall_e/data.py
index b062d60..dd10c56 100755
--- a/vall_e/data.py
+++ b/vall_e/data.py
@@ -185,9 +185,11 @@ class Dataset(_Dataset):
# would be a better cost saving if we could fetch the duration during the validation pass but oh well
self.duration = 0
+ """
if cfg.dataset.use_hdf5:
- for path in self.paths:
+ for path in tqdm(self.paths, desc="Calculating duration"):
self.duration += cfg.hdf5[_get_hdf5_path(path)].attrs['duration']
+ """
@cached_property
def phones(self):
diff --git a/vall_e/models/ar.py b/vall_e/models/ar.py
index 0d370db..c8ac130 100755
--- a/vall_e/models/ar.py
+++ b/vall_e/models/ar.py
@@ -63,7 +63,7 @@ class AR(Base):
max_steps: int = 1000,
sampling_temperature: float = 1.0,
- naive: bool = False,
+ naive: bool = True,
):
if resps_list is not None:
resps_list = [r[..., 0] for r in resps_list] # guarantees we only have the first levels
diff --git a/vall_e/models/base.py b/vall_e/models/base.py
index f149065..3b3f7ff 100755
--- a/vall_e/models/base.py
+++ b/vall_e/models/base.py
@@ -387,7 +387,7 @@ def example_usage():
from .ar import AR
from .nar import NAR
- device = "cpu"
+ device = "cuda"
x8 = partial(repeat, pattern="t -> t l", l=2)
symmap = {'': 1, '': 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}
def tokenize(content, lang_marker="en"):
@@ -444,12 +444,12 @@ def example_usage():
engines.train()
t = trange(60)
for i in t:
- """
stats = {"step": i}
for name, engine in engines.items():
stats |= engine.traverse(text_list=text_list, proms_list=proms_list, resps_list=resps_list)
"""
- stats = engines.step({"text_list": text_list, "proms_list": proms_list, "resps_list": resps_list}, device="cpu")
+ stats = engines.step({"text_list": text_list, "proms_list": proms_list, "resps_list": resps_list})
+ """
t.set_description(f"{stats}")
else:
for name, engine in engines.items():
diff --git a/vall_e/models/nar.py b/vall_e/models/nar.py
index d60a49b..953f419 100755
--- a/vall_e/models/nar.py
+++ b/vall_e/models/nar.py
@@ -127,7 +127,7 @@ def example_usage():
from ..engines import Engine
from tqdm import tqdm
- device = "cpu"
+ device = "cuda"
x8 = partial(repeat, pattern="t -> t l", l=cfg.models.prom_levels)
symmap = {'': 1, '': 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}
def tokenize(content, lang_marker="en"):