Package everything

requirements.txt

@@ -7,4 +7,6 @@ inflect
 progressbar
 einops
 unidecode
-entmax
+entmax
+scipy
+librosa

setup.py

@@ -0,0 +1,22 @@
+from setuptools import setup, find_packages
+
+install_requires = [
+    "torch",
+    "torchaudio",
+    "rotary_embedding_torch",
+    "transformers",
+    "tokenizers",
+    "inflect",
+    "progressbar",
+    "einops",
+    "unidecode",
+    "entmax",
+    "scipy",
+    "librosa"
+]
+
+setup(
+    name="tortoise_tts",
+    packages=['tortoise_tts'],
+    install_requires=install_requires,
+)

tortoise_tts/__init__.py

@@ -0,0 +1 @@
+from .api import TextToSpeech

api.py → tortoise_tts/api.py

@@ -8,18 +8,18 @@ import torch.nn.functional as F
 import progressbar
 import torchaudio
 
-from models.classifier import AudioMiniEncoderWithClassifierHead
-from models.cvvp import CVVP
-from models.diffusion_decoder import DiffusionTts
-from models.autoregressive import UnifiedVoice
+from tortoise_tts.models.classifier import AudioMiniEncoderWithClassifierHead
+from tortoise_tts.models.cvvp import CVVP
+from tortoise_tts.models.diffusion_decoder import DiffusionTts
+from tortoise_tts.models.autoregressive import UnifiedVoice
 from tqdm import tqdm
 
-from models.arch_util import TorchMelSpectrogram
-from models.clvp import CLVP
-from models.vocoder import UnivNetGenerator
-from utils.audio import load_audio, wav_to_univnet_mel, denormalize_tacotron_mel
-from utils.diffusion import SpacedDiffusion, space_timesteps, get_named_beta_schedule
-from utils.tokenizer import VoiceBpeTokenizer, lev_distance
+from tortoise_tts.models.arch_util import TorchMelSpectrogram
+from tortoise_tts.models.clvp import CLVP
+from tortoise_tts.models.vocoder import UnivNetGenerator
+from tortoise_tts.utils.audio import load_audio, wav_to_univnet_mel, denormalize_tacotron_mel
+from tortoise_tts.utils.diffusion import SpacedDiffusion, space_timesteps, get_named_beta_schedule
+from tortoise_tts.utils.tokenizer import VoiceBpeTokenizer, lev_distance
 
 
 pbar = None

{models → tortoise_tts/models}/arch_util.py

@@ -5,7 +5,7 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import torchaudio
-from models.xtransformers import ContinuousTransformerWrapper, RelativePositionBias
+from tortoise_tts.models.xtransformers import ContinuousTransformerWrapper, RelativePositionBias
 
 
 def zero_module(module):

{models → tortoise_tts/models}/autoregressive.py

@@ -6,8 +6,8 @@ import torch.nn.functional as F
 from transformers import GPT2Config, GPT2PreTrainedModel, LogitsProcessorList
 from transformers.modeling_outputs import CausalLMOutputWithCrossAttentions
 from transformers.utils.model_parallel_utils import get_device_map, assert_device_map
-from models.arch_util import AttentionBlock
-from utils.typical_sampling import TypicalLogitsWarper
+from tortoise_tts.models.arch_util import AttentionBlock
+from tortoise_tts.utils.typical_sampling import TypicalLogitsWarper
 
 
 def null_position_embeddings(range, dim):

{models → tortoise_tts/models}/classifier.py

@@ -3,7 +3,7 @@ import torch.nn as nn
 import torch.nn.functional as F
 from torch.utils.checkpoint import checkpoint
 
-from models.arch_util import Upsample, Downsample, normalization, zero_module, AttentionBlock
+from tortoise_tts.models.arch_util import Upsample, Downsample, normalization, zero_module, AttentionBlock
 
 
 class ResBlock(nn.Module):

{models → tortoise_tts/models}/clvp.py

@@ -3,9 +3,9 @@ import torch.nn as nn
 import torch.nn.functional as F
 from torch import einsum
 
-from models.arch_util import CheckpointedXTransformerEncoder
-from models.transformer import Transformer
-from models.xtransformers import Encoder
+from tortoise_tts.models.arch_util import CheckpointedXTransformerEncoder
+from tortoise_tts.models.transformer import Transformer
+from tortoise_tts.models.xtransformers import Encoder
 
 
 def exists(val):

{models → tortoise_tts/models}/cvvp.py

@@ -4,8 +4,8 @@ import torch.nn.functional as F
 from torch import einsum
 from torch.utils.checkpoint import checkpoint
 
-from models.arch_util import AttentionBlock
-from models.xtransformers import ContinuousTransformerWrapper, Encoder
+from tortoise_tts.models.arch_util import AttentionBlock
+from tortoise_tts.models.xtransformers import ContinuousTransformerWrapper, Encoder
 
 
 def exists(val):

{models → tortoise_tts/models}/diffusion_decoder.py

@@ -7,7 +7,7 @@ import torch.nn as nn
 import torch.nn.functional as F
 from torch import autocast
 
-from models.arch_util import normalization, AttentionBlock
+from tortoise_tts.models.arch_util import normalization, AttentionBlock
 
 
 def is_latent(t):

tortoise_tts/utils/__init__.py

@@ -0,0 +1,4 @@
+from .audio import (
+    load_audio,
+    get_voices
+)

{utils → tortoise_tts/utils}/audio.py

@@ -6,7 +6,7 @@ import torchaudio
 import numpy as np
 from scipy.io.wavfile import read
 
-from utils.stft import STFT
+from tortoise_tts.utils.stft import STFT
 
 
 def load_wav_to_torch(full_path):

{utils → tortoise_tts/utils}/tokenizer.py

@@ -3,6 +3,7 @@ import re
 import inflect
 import torch
 from tokenizers import Tokenizer
+from pathlib import Path
 
 
 # Regular expression matching whitespace:
@@ -165,6 +166,8 @@ def lev_distance(s1, s2):
 
 class VoiceBpeTokenizer:
     def __init__(self, vocab_file='data/tokenizer.json'):
+        vocab_file = str(Path(__file__).parent.parent / Path(vocab_file))
+        print(vocab_file)
         if vocab_file is not None:
             self.tokenizer = Tokenizer.from_file(vocab_file)