lt-tensor 0.0.1a17__py3-none-any.whl → 0.0.1a18__py3-none-any.whl

lt_tensor/model_zoo/audio_models/__init__.py

@@ -1,2 +1,3 @@
 from . import diffwave, istft, hifigan
 
+__all__ = ["diffwave", "istft", "hifigan"]

lt_tensor/model_zoo/audio_models/diffwave/__init__.py

@@ -1,4 +1,4 @@
-__all__ = ["DiffWave", "SpectrogramUpsampler", "DiffusionEmbedding"]
+__all__ = ["DiffWave", "DiffWaveConfig", "SpectrogramUpsample", "DiffusionEmbedding"]
 
 import numpy as np
 import torch
@@ -10,8 +10,6 @@ from lt_tensor.model_base import Model
 from math import sqrt
 from lt_utils.common import *
 
-F.t
-
 
 class DiffWaveConfig(ModelConfig):
     # Training params
@@ -40,7 +38,6 @@ class DiffWaveConfig(ModelConfig):
         settings: Dict[str, Any] = {},
         path_name: Optional[Union[str, PathLike]] = None,
     ):
-        self._forbidden_list.extend()
         super().__init__(settings, path_name)
 
 
@@ -86,7 +83,7 @@ class DiffusionEmbedding(Model):
         return table
 
 
-class SpectrogramUpsampler(Model):
+class SpectrogramUpsample(Model):
     def __init__(self):
         super().__init__()
         self.conv1 = nn.ConvTranspose2d(1, 1, [3, 32], stride=[1, 16], padding=[1, 8])
@@ -162,7 +159,7 @@ class DiffWave(Model):
         if self.params.unconditional:  # use unconditional model
             self.spectrogram_upsample = None
         else:
-            self.spectrogram_upsample = SpectrogramUpsampler()
+            self.spectrogram_upsample = SpectrogramUpsample()
 
         self.residual_layers = nn.ModuleList(
             [

lt_tensor/model_zoo/audio_models/hifigan/__init__.py

@@ -1,4 +1,4 @@
-__all__ = ["HifiganGenerator"]
+__all__ = ["HifiganGenerator", "HifiganConfig"]
 from lt_utils.common import *
 from lt_tensor.torch_commons import *
 from lt_tensor.model_zoo.residual import ConvNets
@@ -13,6 +13,33 @@ def get_padding(kernel_size, dilation=1):
     return int((kernel_size * dilation - dilation) / 2)
 
 
+from lt_tensor.config_templates import ModelConfig
+
+
+class HifiganConfig(ModelConfig):
+    # Training params
+    in_channels: int = 80
+    upsample_rates: List[Union[int, List[int]]] = [8, 8]
+    upsample_kernel_sizes: List[Union[int, List[int]]] = [16, 16]
+    upsample_initial_channel: int = (512,)
+    resblock_kernel_sizes: List[Union[int, List[int]]] = [3, 7, 11]
+    resblock_dilation_sizes: List[Union[int, List[int]]] = [
+        [1, 3, 5],
+        [1, 3, 5],
+        [1, 3, 5],
+    ]
+
+    activation: nn.Module = nn.LeakyReLU(0.1)
+    resblock: int = 0
+
+    def __init__(
+        self,
+        settings: Dict[str, Any] = {},
+        path_name: Optional[Union[str, PathLike]] = None,
+    ):
+        super().__init__(settings, path_name)
+
+
 class ResBlock1(ConvNets):
     def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
         super().__init__()
@@ -142,23 +169,23 @@ class ResBlock2(ConvNets):
 
 
 class HifiganGenerator(ConvNets):
-    def __init__(self, h):
+    def __init__(self, cfg: HifiganConfig = HifiganConfig()):
         super().__init__()
-        self.h = h
-        self.num_kernels = len(h.resblock_kernel_sizes)
-        self.num_upsamples = len(h.upsample_rates)
+        self.cfg = cfg
+        self.num_kernels = len(cfg.resblock_kernel_sizes)
+        self.num_upsamples = len(cfg.upsample_rates)
         self.conv_pre = weight_norm(
-            nn.Conv1d(80, h.upsample_initial_channel, 7, 1, padding=3)
+            nn.Conv1d(cfg.in_channels, cfg.upsample_initial_channel, 7, 1, padding=3)
         )
-        resblock = ResBlock1 if h.resblock == "1" else ResBlock2
-        self.activation = nn.LeakyReLU(0.1)
+        resblock = ResBlock1 if resblock == 0 else ResBlock2
+        self.activation = cfg.activation
         self.ups = nn.ModuleList()
-        for i, (u, k) in enumerate(zip(h.upsample_rates, h.upsample_kernel_sizes)):
+        for i, (u, k) in enumerate(zip(cfg.psample_rates, cfg.upsample_kernel_sizes)):
             self.ups.append(
                 weight_norm(
                     nn.ConvTranspose1d(
-                        h.upsample_initial_channel // (2**i),
-                        h.upsample_initial_channel // (2 ** (i + 1)),
+                        cfg.upsample_initial_channel // (2**i),
+                        cfg.upsample_initial_channel // (2 ** (i + 1)),
                         k,
                         u,
                         padding=(k - u) // 2,
@@ -168,17 +195,17 @@ class HifiganGenerator(ConvNets):
 
         self.resblocks = nn.ModuleList()
         for i in range(len(self.ups)):
-            ch = h.upsample_initial_channel // (2 ** (i + 1))
+            ch = cfg.upsample_initial_channel // (2 ** (i + 1))
             for j, (k, d) in enumerate(
-                zip(h.resblock_kernel_sizes, h.resblock_dilation_sizes)
+                zip(cfg.resblock_kernel_sizes, cfg.resblock_dilation_sizes)
             ):
-                self.resblocks.append(resblock(h, ch, k, d))
+                self.resblocks.append(resblock(ch, k, d))
 
         self.conv_post = weight_norm(nn.Conv1d(ch, 1, 7, 1, padding=3))
         self.ups.apply(self.init_weights)
         self.conv_post.apply(self.init_weights)
 
-    def forward(self, x):
+    def forward(self, x: Tensor):
         x = self.conv_pre(x)
         for i in range(self.num_upsamples):
             x = self.ups[i](self.activation(x))

lt_tensor/model_zoo/audio_models/istft/__init__.py

@@ -1,8 +1,35 @@
-__all__ = ["iSTFTGenerator"]
+__all__ = ["iSTFTNetGenerator", "iSTFTNetConfig"]
 from lt_utils.common import *
 from lt_tensor.torch_commons import *
 from lt_tensor.model_zoo.residual import ConvNets
 from torch.nn import functional as F
+from lt_tensor.config_templates import ModelConfig
+
+
+class iSTFTNetConfig(ModelConfig):
+    # Training params
+    in_channels: int = 80
+    upsample_rates: List[Union[int, List[int]]] = [8, 8]
+    upsample_kernel_sizes: List[Union[int, List[int]]] = [16, 16]
+    upsample_initial_channel: int = (512,)
+    resblock_kernel_sizes: List[Union[int, List[int]]] = [3, 7, 11]
+    resblock_dilation_sizes: List[Union[int, List[int]]] = [
+        [1, 3, 5],
+        [1, 3, 5],
+        [1, 3, 5],
+    ]
+
+    activation: nn.Module = nn.LeakyReLU(0.1)
+    resblock: int = 0
+    gen_istft_n_fft: int = 16
+    sampling_rate: Number = 24000
+
+    def __init__(
+        self,
+        settings: Dict[str, Any] = {},
+        path_name: Optional[Union[str, PathLike]] = None,
+    ):
+        super().__init__(settings, path_name)
 
 
 def get_padding(ks, d):
@@ -10,9 +37,8 @@ def get_padding(ks, d):
 
 
 class ResBlock1(ConvNets):
-    def __init__(self, h, channels, kernel_size=3, dilation=(1, 3, 5)):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
         super().__init__()
-        self.h = h
         self.convs1 = nn.ModuleList(
             [
                 weight_norm(
@@ -95,10 +121,10 @@ class ResBlock1(ConvNets):
             x = xt + x
         return x
 
+
 class ResBlock2(ConvNets):
-    def __init__(self, h, channels, kernel_size=3, dilation=(1, 3)):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
         super().__init__()
-        self.h = h
         self.convs = nn.ModuleList(
             [
                 weight_norm(
@@ -134,25 +160,25 @@ class ResBlock2(ConvNets):
         return x
 
 
-class iSTFTGenerator(ConvNets):
-    def __init__(self, h):
+class iSTFTNetGenerator(ConvNets):
+    def __init__(self, cfg: iSTFTNetConfig = iSTFTNetConfig()):
         super().__init__()
-        self.h = h
-        self.num_kernels = len(h.resblock_kernel_sizes)
-        self.num_upsamples = len(h.upsample_rates)
+        self.cfg = cfg
+        self.num_kernels = len(cfg.resblock_kernel_sizes)
+        self.num_upsamples = len(cfg.upsample_rates)
         self.conv_pre = weight_norm(
-            nn.Conv1d(80, h.upsample_initial_channel, 7, 1, padding=3)
+            nn.Conv1d(cfg.in_channels, cfg.upsample_initial_channel, 7, 1, padding=3)
         )
-        resblock = ResBlock1 if h.resblock == "1" else ResBlock2
+        resblock = ResBlock1 if resblock == 0 else ResBlock2
 
         self.ups = nn.ModuleList()
-        for i, (u, k) in enumerate(zip(h.upsample_rates, h.upsample_kernel_sizes)):
-            if h.sampling_rate % 16000:
+        for i, (u, k) in enumerate(zip(cfg.upsample_rates, cfg.upsample_kernel_sizes)):
+            if cfg.sampling_rate % 16000:
                 self.ups.append(
                     weight_norm(
                         nn.ConvTranspose1d(
-                            h.upsample_initial_channel // (2**i),
-                            h.upsample_initial_channel // (2 ** (i + 1)),
+                            cfg.upsample_initial_channel // (2**i),
+                            cfg.upsample_initial_channel // (2 ** (i + 1)),
                             k,
                             u,
                             padding=(k - u) // 2,
@@ -163,8 +189,8 @@ class iSTFTGenerator(ConvNets):
                 self.ups.append(
                     weight_norm(
                         nn.ConvTranspose1d(
-                            h.upsample_initial_channel // (2**i),
-                            h.upsample_initial_channel // (2 ** (i + 1)),
+                            cfg.upsample_initial_channel // (2**i),
+                            cfg.upsample_initial_channel // (2 ** (i + 1)),
                             k,
                             u,
                             padding=(u // 2 + u % 2),
@@ -175,19 +201,19 @@ class iSTFTGenerator(ConvNets):
 
         self.resblocks = nn.ModuleList()
         for i in range(len(self.ups)):
-            ch = h.upsample_initial_channel // (2 ** (i + 1))
+            ch = cfg.upsample_initial_channel // (2 ** (i + 1))
             for j, (k, d) in enumerate(
-                zip(h.resblock_kernel_sizes, h.resblock_dilation_sizes)
+                zip(cfg.resblock_kernel_sizes, cfg.resblock_dilation_sizes)
             ):
-                self.resblocks.append(resblock(h, ch, k, d))
+                self.resblocks.append(resblock(ch, k, d))
 
-        self.post_n_fft = h.gen_istft_n_fft
+        self.post_n_fft = cfg.gen_istft_n_fft
         self.conv_post = weight_norm(
             nn.Conv1d(ch, self.post_n_fft + 2, 7, 1, padding=3)
         )
         self.ups.apply(self.init_weights)
         self.conv_post.apply(self.init_weights)
-        self.activation = nn.LeakyReLU(0.1)
+        self.activation = cfg.activation
         self.reflection_pad = torch.nn.ReflectionPad1d((1, 0))
 
     def forward(self, x):

{lt_tensor-0.0.1a17.dist-info → lt_tensor-0.0.1a18.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lt-tensor
-Version: 0.0.1a17
+Version: 0.0.1a18
 Summary: General utilities for PyTorch and others. Built for general use.
 Home-page: https://github.com/gr1336/lt-tensor/
 Author: gr1336

{lt_tensor-0.0.1a17.dist-info → lt_tensor-0.0.1a18.dist-info}/RECORD

@@ -18,14 +18,14 @@ lt_tensor/model_zoo/fusion.py,sha256=usC1bcjQRNivDc8xzkIS5T1glm78OLcs2V_tPqfp-eI
 lt_tensor/model_zoo/pos_encoder.py,sha256=3d1EYLinCU9UAy-WuEWeYMGhMqaGknCiQ5qEmhw_UYM,4487
 lt_tensor/model_zoo/residual.py,sha256=i5V4ju7DB3WesKBVm6KH_LyPoKGDUOyo2Usfs-PyP58,9394
 lt_tensor/model_zoo/transformer.py,sha256=HUFoFFh7EQJErxdd9XIxhssdjvNVx2tNGDJOTUfwG2A,4301
-lt_tensor/model_zoo/audio_models/__init__.py,sha256=CmoakfBLoxqtJuYc1NYrB_z0x1kS2WQNaYQRmCaC5ko,40
-lt_tensor/model_zoo/audio_models/diffwave/__init__.py,sha256=8DbKJpQ44s9iPlajfs7_A2N1diYGXzkhet_wS4hX6mU,7421
-lt_tensor/model_zoo/audio_models/hifigan/__init__.py,sha256=BOBZSK2HFOdMcFyjrzwZi_TeAtBGIcpb8pQxiGlwLEE,12302
-lt_tensor/model_zoo/audio_models/istft/__init__.py,sha256=o7Ie1qI22u_g9t1252PX4vl4uF6JHynAJryuz2lAZE0,12920
+lt_tensor/model_zoo/audio_models/__init__.py,sha256=MoG9YjxLyvscq_6njK1ljGBletK9iedBXt66bplzW-s,83
+lt_tensor/model_zoo/audio_models/diffwave/__init__.py,sha256=R14hY-nCbCO-T3ox9f4MXCPgQQogFUKAJ2WtntLz09w,7393
+lt_tensor/model_zoo/audio_models/hifigan/__init__.py,sha256=6ZGYyNiTMGHnOjGU0gq_TSM8Y9LtYlP3neGwa01Ghyk,13135
+lt_tensor/model_zoo/audio_models/istft/__init__.py,sha256=noi4GLGZQ_qg5H-ipe5d7j8rvt4Hic_sXiME-TE-B2c,13783
 lt_tensor/processors/__init__.py,sha256=4b9MxAJolXiJfSm20ZEspQTDm1tgLazwlPWA_jB1yLM,63
 lt_tensor/processors/audio.py,sha256=SMqNSl4Den-x1awTCQ8-TcR-0jPiv5lDaUpU93SRRaw,14749
-lt_tensor-0.0.1a17.dist-info/licenses/LICENSE,sha256=HUnu_iSPpnDfZS_PINhO3AoVizJD1A2vee8WX7D7uXo,11358
-lt_tensor-0.0.1a17.dist-info/METADATA,sha256=nHIAMKShjCnhr2KdEiAhi8IIqP9PoTWnGHgsSfNcTDs,1033
-lt_tensor-0.0.1a17.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lt_tensor-0.0.1a17.dist-info/top_level.txt,sha256=35FuhFeXnUyvHWdbVHGPh0hS8euofafnJ_GJAVSF4Kk,10
-lt_tensor-0.0.1a17.dist-info/RECORD,,
+lt_tensor-0.0.1a18.dist-info/licenses/LICENSE,sha256=HUnu_iSPpnDfZS_PINhO3AoVizJD1A2vee8WX7D7uXo,11358
+lt_tensor-0.0.1a18.dist-info/METADATA,sha256=fgRzOiw5tMmkaEY9HrGEKNL2v9mN5JVbf9r-bf18Am0,1033
+lt_tensor-0.0.1a18.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lt_tensor-0.0.1a18.dist-info/top_level.txt,sha256=35FuhFeXnUyvHWdbVHGPh0hS8euofafnJ_GJAVSF4Kk,10
+lt_tensor-0.0.1a18.dist-info/RECORD,,