mizoru

from fish_speech.models.text2semantic.inference import *

text: str  = ["Что? Он никогда не работал?", "Никогда бы о таком не подумал раньше"]
prompt_text: Optional[list[str]]  = ["Что он никогда вот не работал, но стоит ему где-то пристроиться, он тут же идёт в гору очень быстро"] * 2
prompt_tokens: Optional[list[Path]]  = [Path("prompt_ORD_1.npy"), ] * 2
num_samples: int  = 1
max_new_tokens: int  = 256
top_p: int  = 0.7
repetition_penalty: float  = 1.2
temperature: float  = 0.7

mizoru

cd ~ && find . -name "*:Zone.Identifier" -type f -delete && cd - > /dev/null

mizoru

x = torch.ones(bs, 784) * 0.5 + 0.02 * torch.randn(bs, 784)
x.requires_grad_(True).retain_grad()
for i in range(0, n//bs):
    preds = model(x)
    loss = loss_func(preds, y)
    loss.backward()
    with torch.no_grad():
        x -= x.grad * lr
        if x.grad.grad_fn is not None:
            x.grad.detach_()

mizoru

def get_data_to_buffer(train_config):
    buffer = list()
    text = process_text(train_config.data_path)
    audio_files = sorted(Path(train_config.audio_path).iterdir())
    spec = Spectrogram(512)

    start = time.perf_counter()
    for i, file in tqdm(zip(range(len(text)), audio_files)):

        mel_gt_name = os.path.join(

mizoru

def get_data_to_buffer(train_config):
    buffer = list()
    text = process_text(train_config.data_path)
    audio_files = sorted(Path(train_config.audio_path).iterdir())
    hop_length = 256
    frame_period = hop_length / 22_050 * 1000

    start = time.perf_counter()
    for i, file in tqdm(zip(range(len(text)), audio_files)):

mizoru

CORR = {'h#': "",
 'sh': "ʃ",
 'iy': "i",
 'hv': "ɦ",
 'ae': "æ",
 'dcl': "d̚",
 'd': "d",
 'y': "j",
 'er': "ɝ",
 'aa': "ɑ",

mizoru

def replace_slashes(s:str):
    return s.replace('\\', '/')
  
df.path = df.path.map(replace_slashes)