:albemic: | More tests.

.gitignore

@@ -166,3 +166,4 @@ dataset/
 old-output/
 output/
 *.wav
+models/

AudioUtils.py

@@ -0,0 +1,18 @@
+import torch
+import torch.nn.functional as F
+
+def stereo_tensor_to_mono(waveform):
+    if waveform.shape[0] > 1:
+        # Average across channels
+        mono_waveform = torch.mean(waveform, dim=0, keepdim=True)
+    else:
+        # Already mono
+        mono_waveform = waveform
+    return mono_waveform
+
+def stretch_tensor(tensor, target_length):
+    scale_factor = target_length / tensor.size(1)
+
+    tensor = F.interpolate(tensor, scale_factor=scale_factor, mode='linear', align_corners=False)
+
+    return tensor

data.py

@@ -1,50 +1,53 @@
 from torch.utils.data import Dataset
 import torch.nn.functional as F
+import torch
 import torchaudio
 import os
 import random
-
+import torchaudio.transforms as T
+import AudioUtils
 
 class AudioDataset(Dataset):
-    audio_sample_rates = [8000, 11025, 16000, 22050]
+    audio_sample_rates = [11025]
+    MAX_LENGTH = 44100  # Define your desired maximum length here
 
-    def __init__(self, input_dir, target_duration=None, padding_mode='constant', padding_value=0.0):
+    def __init__(self, input_dir, device):
-        self.input_files = [os.path.join(input_dir, f) for f in os.listdir(input_dir) if f.endswith('.wav')]
+        self.input_files = [os.path.join(root, f) for root, _, files in os.walk(input_dir) for f in files if f.endswith('.wav')]
-        self.target_duration = target_duration  # Duration in seconds or None if not set
+        self.device = device
-        self.padding_mode = padding_mode
-        self.padding_value = padding_value
 
     def __len__(self):
         return len(self.input_files)
 
-
     def __getitem__(self, idx):
-        high_quality_wav, sr_original = torchaudio.load(self.input_files[idx], normalize=True)
+        # Load high-quality audio
+        high_quality_audio, original_sample_rate = torchaudio.load(self.input_files[idx], normalize=True)
 
-        sample_rate = random.choice(self.audio_sample_rates)
+        # Generate low-quality audio with random downsampling
-        resample_transform = torchaudio.transforms.Resample(sr_original, sample_rate)
+        mangled_sample_rate = random.choice(self.audio_sample_rates)
-        low_quality_wav = resample_transform(high_quality_wav)
+        resample_transform_low = torchaudio.transforms.Resample(original_sample_rate, mangled_sample_rate)
-        low_quality_wav = low_quality_wav
+        low_quality_audio = resample_transform_low(high_quality_audio)
 
-        # Calculate target length based on desired duration and 16000 Hz
+        resample_transform_high = torchaudio.transforms.Resample(mangled_sample_rate, original_sample_rate)
-        if self.target_duration is not None:
+        low_quality_audio = resample_transform_high(low_quality_audio)
-            target_length = int(self.target_duration * 44100)
-        else:
-            # Calculate duration of original high quality audio
-            target_length = high_quality_wav.size(1)
 
-        # Pad both to the calculated target length
+        high_quality_audio = AudioUtils.stereo_tensor_to_mono(high_quality_audio)
-        high_quality_wav = self.stretch_tensor(high_quality_wav, target_length)
+        low_quality_audio = AudioUtils.stereo_tensor_to_mono(low_quality_audio)
-        low_quality_wav = self.stretch_tensor(low_quality_wav, target_length)
 
+        # Pad or truncate high-quality audio
+        if high_quality_audio.shape[1] < self.MAX_LENGTH:
+            padding = self.MAX_LENGTH - high_quality_audio.shape[1]
+            high_quality_audio = F.pad(high_quality_audio, (0, padding))
+        elif high_quality_audio.shape[1] > self.MAX_LENGTH:
+            high_quality_audio = high_quality_audio[:, :self.MAX_LENGTH]
 
-        return low_quality_wav, high_quality_wav
+        # Pad or truncate low-quality audio
+        if low_quality_audio.shape[1] < self.MAX_LENGTH:
+            padding = self.MAX_LENGTH - low_quality_audio.shape[1]
+            low_quality_audio = F.pad(low_quality_audio, (0, padding))
+        elif low_quality_audio.shape[1] > self.MAX_LENGTH:
+            low_quality_audio = low_quality_audio[:, :self.MAX_LENGTH]
 
-    def stretch_tensor(self, tensor, target_length):
+        high_quality_audio = high_quality_audio.to(self.device)
-        current_length = tensor.size(1)
+        low_quality_audio = low_quality_audio.to(self.device)
-        scale_factor = target_length / current_length
 
-        # Resample the tensor using linear interpolation
+        return (high_quality_audio, original_sample_rate), (low_quality_audio, mangled_sample_rate)
-        tensor = F.interpolate(tensor.unsqueeze(0), scale_factor=scale_factor, mode='linear', align_corners=False).squeeze(0)
-
-        return tensor

discriminator.py

@@ -1,24 +1,58 @@
+import torch
 import torch.nn as nn
+import torch.nn.utils as utils
+
+def discriminator_block(in_channels, out_channels, kernel_size=3, stride=1, dilation=1, spectral_norm=True):
+    padding = (kernel_size // 2) * dilation
+    conv_layer = nn.Conv1d(in_channels, out_channels, kernel_size=kernel_size, stride=stride, dilation=dilation, padding=padding)
+    if spectral_norm:
+        conv_layer = utils.spectral_norm(conv_layer)
+    return nn.Sequential(
+        conv_layer,
+        nn.LeakyReLU(0.2, inplace=True),
+        nn.BatchNorm1d(out_channels)
+    )
+
+class AttentionBlock(nn.Module):
+    def __init__(self, channels):
+        super(AttentionBlock, self).__init__()
+        self.attention = nn.Sequential(
+            nn.Conv1d(channels, channels // 4, kernel_size=1),
+            nn.ReLU(),
+            nn.Conv1d(channels // 4, channels, kernel_size=1),
+            nn.Sigmoid()
+        )
+
+    def forward(self, x):
+        attention_weights = self.attention(x)
+        return x * attention_weights
 
 class SISUDiscriminator(nn.Module):
-    def __init__(self):
+    def __init__(self, layers=4): #Increased base layer count
         super(SISUDiscriminator, self).__init__()
         self.model = nn.Sequential(
-            nn.Conv1d(2, 128, kernel_size=3, padding=1),
+            discriminator_block(1, layers, kernel_size=3, stride=1), #Aggressive downsampling
-            nn.LeakyReLU(0.2, inplace=True),
+            discriminator_block(layers, layers * 2, kernel_size=5, stride=2),
-            nn.Conv1d(128, 256, kernel_size=3, padding=1),
+            discriminator_block(layers * 2, layers * 4, kernel_size=5, dilation=4),
-            nn.LeakyReLU(0.2, inplace=True),
+
-            nn.Conv1d(256, 128, kernel_size=3, padding=1),
+            #AttentionBlock(layers * 4), #Added attention
-            nn.LeakyReLU(0.2, inplace=True),
+
-            nn.Conv1d(128, 64, kernel_size=3, padding=1),
+            #discriminator_block(layers * 4, layers * 8, kernel_size=5, dilation=4),
-            nn.LeakyReLU(0.2, inplace=True),
+            #AttentionBlock(layers * 8), #Added attention
-            nn.Conv1d(64, 1, kernel_size=3, padding=1),
+            #discriminator_block(layers * 8, layers * 16, kernel_size=5, dilation=8),
-            nn.LeakyReLU(0.2, inplace=True),
+            #discriminator_block(layers * 16, layers * 16, kernel_size=3, dilation=1),
+            #discriminator_block(layers * 16, layers * 8, kernel_size=3, dilation=2),
+            #discriminator_block(layers * 8, layers * 4, kernel_size=3, dilation=1),
+            discriminator_block(layers * 4, layers * 2, kernel_size=5, stride=2),
+            discriminator_block(layers * 2, layers, kernel_size=3, stride=1),
+            discriminator_block(layers, 1, kernel_size=3, stride=1, spectral_norm=False) #last layer no spectral norm.
         )
-        self.global_avg_pool = nn.AdaptiveAvgPool1d(1)  # Output size (1,)
+        self.global_avg_pool = nn.AdaptiveAvgPool1d(1)
+        self.sigmoid = nn.Sigmoid()
 
     def forward(self, x):
         x = self.model(x)
         x = self.global_avg_pool(x)
-        x = x.view(-1, 1)  # Flatten to (batch_size, 1)
+        x = x.view(-1, 1)
+        x = self.sigmoid(x)
         return x

generator.py

@@ -1,23 +1,52 @@
 import torch.nn as nn
 
-class SISUGenerator(nn.Module):
+def conv_block(in_channels, out_channels, kernel_size=3, dilation=1):
-    def __init__(self, upscale_scale=1): # No noise_dim parameter
+    return nn.Sequential(
-        super(SISUGenerator, self).__init__()
+        nn.Conv1d(in_channels, out_channels, kernel_size=kernel_size, dilation=dilation, padding=(kernel_size // 2) * dilation),
-        self.model = nn.Sequential(
+        nn.BatchNorm1d(out_channels),
-            nn.Conv1d(2, 128, kernel_size=3, padding=1),
+        nn.PReLU()
-            nn.LeakyReLU(0.2, inplace=True),
+    )
-            nn.Conv1d(128, 256, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
 
-            nn.Upsample(scale_factor=upscale_scale, mode='nearest'),
+class AttentionBlock(nn.Module):
-
+    def __init__(self, channels):
-            nn.Conv1d(256, 128, kernel_size=3, padding=1),
+        super(AttentionBlock, self).__init__()
-            nn.LeakyReLU(0.2, inplace=True),
+        self.attention = nn.Sequential(
-            nn.Conv1d(128, 64, kernel_size=3, padding=1),
+            nn.Conv1d(channels, channels // 4, kernel_size=1),
-            nn.LeakyReLU(0.2, inplace=True),
+            nn.ReLU(),
-            nn.Conv1d(64, 2, kernel_size=3, padding=1),
+            nn.Conv1d(channels // 4, channels, kernel_size=1),
-            nn.Tanh()
+            nn.Sigmoid()
         )
 
     def forward(self, x):
-        return self.model(x)
+        attention_weights = self.attention(x)
+        return x * attention_weights
+
+class ResidualInResidualBlock(nn.Module):
+    def __init__(self, channels, num_convs=3):
+        super(ResidualInResidualBlock, self).__init__()
+        self.conv_layers = nn.Sequential(*[conv_block(channels, channels) for _ in range(num_convs)])
+        self.attention = AttentionBlock(channels)
+
+    def forward(self, x):
+        residual = x
+        x = self.conv_layers(x)
+        x = self.attention(x)
+        return x + residual
+
+class SISUGenerator(nn.Module):
+    def __init__(self, layer=4, num_rirb=4): #increased base layer and rirb amounts
+        super(SISUGenerator, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.Conv1d(1, layer, kernel_size=7, padding=3),
+            nn.BatchNorm1d(layer),
+            nn.PReLU(),
+        )
+        self.rir_blocks = nn.Sequential(*[ResidualInResidualBlock(layer) for _ in range(num_rirb)])
+        self.final_layer = nn.Conv1d(layer, 1, kernel_size=3, padding=1)
+
+    def forward(self, x):
+        residual = x
+        x = self.conv1(x)
+        x = self.rir_blocks(x)
+        x = self.final_layer(x)
+        return x + residual

requirements.txt

@@ -1,12 +1,14 @@
-filelock>=3.16.1
+filelock==3.16.1
-fsspec>=2024.10.0
+fsspec==2024.10.0
-Jinja2>=3.1.4
+Jinja2==3.1.4
-MarkupSafe>=2.1.5
+MarkupSafe==2.1.5
-mpmath>=1.3.0
+mpmath==1.3.0
-networkx>=3.4.2
+networkx==3.4.2
-numpy>=2.1.2
+numpy==2.2.3
-pillow>=11.0.0
+pytorch-triton-rocm==3.2.0+git4b3bb1f8
-setuptools>=70.2.0
+setuptools==70.2.0
-sympy>=1.13.1
+sympy==1.13.3
-tqdm>=4.67.1
+torch==2.7.0.dev20250226+rocm6.3
-typing_extensions>=4.12.2
+torchaudio==2.6.0.dev20250226+rocm6.3
+tqdm==4.67.1
+typing_extensions==4.12.2

test.py

@@ -1,10 +0,0 @@
-import torch.nn as nn
-import torch
-from discriminator import SISUDiscriminator
-
-
-discriminator = SISUDiscriminator()
-test_input = torch.randn(1, 2, 1000) # Example input (batch_size, channels, frames)
-output = discriminator(test_input)
-print(output)
-print("Output shape:", output.shape)

training.py

@@ -6,40 +6,120 @@ import torch.nn.functional as F
 import torchaudio
 import tqdm
 
+import argparse
+
+import math
+
+import os
+
 from torch.utils.data import random_split
 from torch.utils.data import DataLoader
 
+import AudioUtils
 from data import AudioDataset
 from generator import SISUGenerator
 from discriminator import SISUDiscriminator
 
-# Check for CUDA availability
+import torchaudio.transforms as T
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# Init script argument parser
+parser = argparse.ArgumentParser(description="Training script")
+parser.add_argument("--generator", type=str, default=None,
+                    help="Path to the generator model file")
+parser.add_argument("--discriminator", type=str, default=None,
+                    help="Path to the discriminator model file")
+parser.add_argument("--device", type=str, default="cpu",  help="Select device")
+parser.add_argument("--epoch", type=int, default=0, help="Current epoch for model versioning")
+parser.add_argument("--debug", action="store_true",  help="Print debug logs")
+
+args = parser.parse_args()
+
+device = torch.device(args.device if torch.cuda.is_available() else "cpu")
 print(f"Using device: {device}")
 
+mfcc_transform = T.MFCC(
+    sample_rate=44100,
+    n_mfcc=20,
+    melkwargs={'n_fft': 2048, 'hop_length': 256}
+).to(device)
+
+def gpu_mfcc_loss(y_true, y_pred):
+    mfccs_true = mfcc_transform(y_true)
+    mfccs_pred = mfcc_transform(y_pred)
+    min_len = min(mfccs_true.shape[2], mfccs_pred.shape[2])
+    mfccs_true = mfccs_true[:, :, :min_len]
+    mfccs_pred = mfccs_pred[:, :, :min_len]
+    loss = torch.mean((mfccs_true - mfccs_pred)**2)
+    return loss
+
+def discriminator_train(high_quality, low_quality, real_labels, fake_labels):
+    optimizer_d.zero_grad()
+
+    # Forward pass for real samples
+    discriminator_decision_from_real = discriminator(high_quality[0])
+    d_loss_real = criterion_d(discriminator_decision_from_real, real_labels)
+
+    # Forward pass for fake samples (from generator output)
+    generator_output = generator(low_quality[0])
+    discriminator_decision_from_fake = discriminator(generator_output.detach())
+    d_loss_fake = criterion_d(discriminator_decision_from_fake, fake_labels)
+
+    # Combine real and fake losses
+    d_loss = (d_loss_real + d_loss_fake) / 2.0
+
+    # Backward pass and optimization
+    d_loss.backward()
+    nn.utils.clip_grad_norm_(discriminator.parameters(), max_norm=1.0)  # Gradient Clipping
+    optimizer_d.step()
+
+    return d_loss
+
+def generator_train(low_quality, high_quality, real_labels):
+    optimizer_g.zero_grad()
+
+    # Forward pass for fake samples (from generator output)
+    generator_output = generator(low_quality[0])
+
+    #mfcc_l = gpu_mfcc_loss(high_quality[0], generator_output)
+
+    discriminator_decision = discriminator(generator_output)
+    adversarial_loss = criterion_g(discriminator_decision, real_labels)
+
+    #combined_loss = adversarial_loss + 0.5 * mfcc_l
+
+    adversarial_loss.backward()
+    optimizer_g.step()
+
+    #return (generator_output, combined_loss, adversarial_loss, mfcc_l)
+    return (generator_output, adversarial_loss)
+
+debug = args.debug
+
 # Initialize dataset and dataloader
 dataset_dir = './dataset/good'
-dataset = AudioDataset(dataset_dir, target_duration=2.0)
+dataset = AudioDataset(dataset_dir, device)
 
-dataset_size = len(dataset)
+# ========= SINGLE =========
-train_size = int(dataset_size * .9)
-val_size = int(dataset_size-train_size)
 
-train_dataset, val_dataset = random_split(dataset, [train_size, val_size])
+train_data_loader = DataLoader(dataset, batch_size=256, shuffle=True)
-
-train_data_loader = DataLoader(train_dataset, batch_size=8, shuffle=True)
-val_data_loader = DataLoader(val_dataset, batch_size=8, shuffle=True)
 
 # Initialize models and move them to device
 generator = SISUGenerator()
 discriminator = SISUDiscriminator()
 
+epoch: int = args.epoch
+
 generator = generator.to(device)
 discriminator = discriminator.to(device)
 
+if args.generator is not None:
+    generator.load_state_dict(torch.load(args.generator, map_location=device, weights_only=True))
+if args.discriminator is not None:
+    discriminator.load_state_dict(torch.load(args.discriminator, map_location=device, weights_only=True))
+
 # Loss
-criterion_g = nn.L1Loss()
+criterion_g = nn.MSELoss()
-criterion_d = nn.BCEWithLogitsLoss()
+criterion_d = nn.BCELoss()
 
 # Optimizers
 optimizer_g = optim.Adam(generator.parameters(), lr=0.0001, betas=(0.5, 0.999))
@@ -49,87 +129,64 @@ optimizer_d = optim.Adam(discriminator.parameters(), lr=0.0001, betas=(0.5, 0.99
 scheduler_g = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer_g, mode='min', factor=0.5, patience=5)
 scheduler_d = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer_d, mode='min', factor=0.5, patience=5)
 
-def snr(y_true, y_pred):
+models_dir = "models"
-    noise = y_true - y_pred
+os.makedirs(models_dir, exist_ok=True)
-    signal_power = torch.mean(y_true ** 2)
-    noise_power = torch.mean(noise ** 2)
-    snr_db = 10 * torch.log10(signal_power / noise_power)
-    return snr_db
-
-def discriminator_train(discriminator, optimizer, criterion, generator, real_labels, fake_labels, high_quality, low_quality):
-    optimizer.zero_grad()
-
-    discriminator_decision_from_real = discriminator(high_quality)
-    d_loss_real = criterion(discriminator_decision_from_real, real_labels)
-
-    generator_output = generator(low_quality)
-    discriminator_decision_from_fake = discriminator(generator_output.detach())
-    d_loss_fake = criterion(discriminator_decision_from_fake, fake_labels)
-
-    d_loss = (d_loss_real + d_loss_fake) / 2.0
-
-    d_loss.backward()
-    nn.utils.clip_grad_norm_(discriminator.parameters(), max_norm=1.0) #Gradient Clipping
-    optimizer.step()
-    # print(f"Discriminator Loss: {d_loss.item():.4f}, Mean Real Logit: {discriminator_decision_from_real.mean().item():.2f}, Mean Fake Logit: {discriminator_decision_from_fake.mean().item():.2f}")
 
 def start_training():
-
+    generator_epochs = 5000
-    # Training loop
-    # discriminator_epochs = 1000
-    generator_epochs = 500
     for generator_epoch in range(generator_epochs):
-        low_quality_audio = torch.empty((1))
+        low_quality_audio = (torch.empty((1)), 1)
-        high_quality_audio = torch.empty((1))
+        high_quality_audio = (torch.empty((1)), 1)
-        ai_enhanced_audio = torch.empty((1))
+        ai_enhanced_audio = (torch.empty((1)), 1)
 
-        # Training
+        times_correct = 0
-        for low_quality, high_quality in tqdm.tqdm(train_data_loader, desc=f"Epoch {generator_epoch+1}/{generator_epochs}"):
-            high_quality = high_quality.to(device)
-            low_quality = low_quality.to(device)
 
-            batch_size = high_quality.size(0)
+        # ========= TRAINING =========
+        for high_quality_clip, low_quality_clip in tqdm.tqdm(train_data_loader, desc=f"Training epoch {generator_epoch+1}/{generator_epochs}, Current epoch {epoch+1}"):
+        # for high_quality_clip, low_quality_clip in train_data_loader:
+            high_quality_sample = (high_quality_clip[0], high_quality_clip[1])
+            low_quality_sample = (low_quality_clip[0], low_quality_clip[1])
+
+            # ========= LABELS =========
+            batch_size = high_quality_clip[0].size(0)
             real_labels = torch.ones(batch_size, 1).to(device)
             fake_labels = torch.zeros(batch_size, 1).to(device)
 
-            # Train Discriminator
+            # ========= DISCRIMINATOR =========
             discriminator.train()
+            d_loss = discriminator_train(high_quality_sample, low_quality_sample, real_labels, fake_labels)
 
-            for _ in range(3):
+            # ========= GENERATOR =========
-                discriminator_train(discriminator, optimizer_d, criterion_d, generator, real_labels, fake_labels, high_quality, low_quality)
-
-            # Train Generator
             generator.train()
-            optimizer_g.zero_grad()
+            #generator_output, combined_loss, adversarial_loss, mfcc_l = generator_train(low_quality_sample, high_quality_sample, real_labels)
+            generator_output, adversarial_loss = generator_train(low_quality_sample, high_quality_sample, real_labels)
 
-            # Generator loss: how well fake data fools the discriminator
+            if debug:
-            generator_output = generator(low_quality)
+                print(d_loss, adversarial_loss)
-            discriminator_decision = discriminator(generator_output)  # No detach here
+            scheduler_d.step(d_loss)
-            g_loss = criterion_g(discriminator_decision, real_labels)  # Train generator to produce real-like outputs
+            scheduler_g.step(adversarial_loss)
 
-            g_loss.backward()
+            # ========= SAVE LATEST AUDIO =========
-            optimizer_g.step()
+            high_quality_audio = (high_quality_clip[0][0], high_quality_clip[1][0])
+            low_quality_audio = (low_quality_clip[0][0], low_quality_clip[1][0])
+            ai_enhanced_audio = (generator_output[0], high_quality_clip[1][0])
 
-            low_quality_audio = low_quality
+        new_epoch = generator_epoch+epoch
-            high_quality_audio = high_quality
-            ai_enhanced_audio = generator_output
-
-        metric = snr(high_quality_audio, ai_enhanced_audio)
-        print(f"Generator metric {metric}!")
-        scheduler_g.step(metric)
 
         if generator_epoch % 10 == 0:
-            print(f"Saved epoch {generator_epoch}!")
+            print(f"Saved epoch {new_epoch}!")
-            torchaudio.save(f"./output/epoch-{generator_epoch}-audio-crap.wav", low_quality_audio[0].cpu(), 44100)
+            torchaudio.save(f"./output/epoch-{new_epoch}-audio-crap.wav", low_quality_audio[0].cpu(), high_quality_audio[1]) # <-- Because audio clip was resampled in data.py from original to crap and to original again.
-            torchaudio.save(f"./output/epoch-{generator_epoch}-audio-ai.wav", ai_enhanced_audio[0].cpu(), 44100)
+            torchaudio.save(f"./output/epoch-{new_epoch}-audio-ai.wav", ai_enhanced_audio[0].cpu(), ai_enhanced_audio[1])
-            torchaudio.save(f"./output/epoch-{generator_epoch}-audio-orig.wav", high_quality_audio[0].cpu(), 44100)
+            torchaudio.save(f"./output/epoch-{new_epoch}-audio-orig.wav", high_quality_audio[0].cpu(), high_quality_audio[1])
 
-        if generator_epoch % 50 == 0:
+        if debug:
-            torch.save(discriminator.state_dict(), "discriminator.pt")
+            print(generator.state_dict().keys())
-            torch.save(generator.state_dict(), "generator.pt")
+            print(discriminator.state_dict().keys())
+        torch.save(discriminator.state_dict(), f"{models_dir}/discriminator_epoch_{new_epoch}.pt")
+        torch.save(generator.state_dict(), f"{models_dir}/generator_epoch_{new_epoch}.pt")
 
-    torch.save(discriminator.state_dict(), "discriminator.pt")
+    torch.save(discriminator, "models/epoch-5000-discriminator.pt")
-    torch.save(generator.state_dict(), "generator.pt")
+    torch.save(generator, "models/epoch-5000-generator.pt")
     print("Training complete!")
 
 start_training()