⚗️ | Added some stupid ways for training + some makeup

.gitignore

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

AudioUtils.py

@@ -0,0 +1,97 @@
+import torch
+import torch.nn.functional as F
+
+
+def stereo_tensor_to_mono(waveform: torch.Tensor) -> torch.Tensor:
+    """
+    Convert stereo (C, N) to mono (1, N). Ensures a channel dimension.
+    """
+    if waveform.dim() == 1:
+        waveform = waveform.unsqueeze(0)  # (N,) -> (1, N)
+
+    if waveform.shape[0] > 1:
+        mono_waveform = torch.mean(waveform, dim=0, keepdim=True)  # (1, N)
+    else:
+        mono_waveform = waveform
+    return mono_waveform
+
+
+def stretch_tensor(tensor: torch.Tensor, target_length: int) -> torch.Tensor:
+    """
+    Stretch audio along time dimension to target_length.
+    Input assumed (1, N). Returns (1, target_length).
+    """
+    if tensor.dim() == 1:
+        tensor = tensor.unsqueeze(0)  # ensure (1, N)
+
+    tensor = tensor.unsqueeze(0)  # (1, 1, N) for interpolate
+    stretched = F.interpolate(
+        tensor, size=target_length, mode="linear", align_corners=False
+    )
+    return stretched.squeeze(0)  # back to (1, target_length)
+
+
+def pad_tensor(audio_tensor: torch.Tensor, target_length: int = 128) -> torch.Tensor:
+    """
+    Pad to fixed length. Input assumed (1, N). Returns (1, target_length).
+    """
+    if audio_tensor.dim() == 1:
+        audio_tensor = audio_tensor.unsqueeze(0)
+
+    current_length = audio_tensor.shape[-1]
+    if current_length < target_length:
+        padding_needed = target_length - current_length
+        padding_tuple = (0, padding_needed)
+        padded_audio_tensor = F.pad(
+            audio_tensor, padding_tuple, mode="constant", value=0
+        )
+    else:
+        padded_audio_tensor = audio_tensor[..., :target_length]  # crop if too long
+
+    return padded_audio_tensor
+
+
+def split_audio(
+    audio_tensor: torch.Tensor, chunk_size: int = 128
+) -> list[torch.Tensor]:
+    """
+    Split into chunks of (1, chunk_size).
+    """
+    if not isinstance(chunk_size, int) or chunk_size <= 0:
+        raise ValueError("chunk_size must be a positive integer.")
+
+    if audio_tensor.dim() == 1:
+        audio_tensor = audio_tensor.unsqueeze(0)
+
+    num_samples = audio_tensor.shape[-1]
+    if num_samples == 0:
+        return []
+
+    chunks = list(torch.split(audio_tensor, chunk_size, dim=-1))
+    return chunks
+
+
+def reconstruct_audio(chunks: list[torch.Tensor]) -> torch.Tensor:
+    """
+    Reconstruct audio from chunks. Returns (1, N).
+    """
+    if not chunks:
+        return torch.empty(1, 0)
+
+    chunks = [c if c.dim() == 2 else c.unsqueeze(0) for c in chunks]
+    try:
+        reconstructed_tensor = torch.cat(chunks, dim=-1)
+    except RuntimeError as e:
+        raise RuntimeError(
+            f"Failed to concatenate audio chunks. Ensure chunks have compatible shapes "
+            f"for concatenation along dim -1. Original error: {e}"
+        )
+
+    return reconstructed_tensor
+
+
+def normalize(audio_tensor: torch.Tensor, eps: float = 1e-8) -> torch.Tensor:
+    max_val = torch.max(torch.abs(audio_tensor))
+    if max_val < eps:
+        return audio_tensor  # silence, skip normalization
+    return audio_tensor / max_val

README.md

@@ -18,6 +18,7 @@ SISU (Super Ingenious Sound Upscaler) is a project that uses GANs (Generative Ad
 1. **Set Up**:
    - Make sure you have Python installed (version 3.8 or higher).
    - Install needed packages: `pip install -r requirements.txt`
+   - Install current version of PyTorch (CUDA/ROCm/What ever your device supports)
 
 2. **Prepare Audio Data**:
    - Put your audio files in the `dataset/good` folder.

app.py

@@ -0,0 +1,97 @@
+import argparse
+
+import torch
+import torchaudio
+import torchcodec
+import tqdm
+
+import AudioUtils
+from generator import SISUGenerator
+
+# Init script argument parser
+parser = argparse.ArgumentParser(description="Training script")
+parser.add_argument("--device", type=str, default="cpu", help="Select device")
+parser.add_argument("--model", type=str, help="Model to use for upscaling")
+parser.add_argument(
+    "--clip_length",
+    type=int,
+    default=16384,
+    help="Internal clip length, leave unspecified if unsure",
+)
+parser.add_argument(
+    "--sample_rate", type=int, default=44100, help="Output clip sample rate"
+)
+parser.add_argument(
+    "--bitrate",
+    type=int,
+    default=192000,
+    help="Output clip bitrate",
+)
+parser.add_argument("-i", "--input", type=str, help="Input audio file")
+parser.add_argument("-o", "--output", type=str, help="Output audio file")
+
+args = parser.parse_args()
+
+if args.sample_rate < 8000:
+    print(
+        "Sample rate cannot be lower than 8000! (44100 is recommended for base models)"
+    )
+    exit()
+
+device = torch.device(args.device if torch.cuda.is_available() else "cpu")
+print(f"Using device: {device}")
+
+generator = SISUGenerator().to(device)
+
+generator = torch.compile(generator)
+
+models_dir = args.model
+clip_length = args.clip_length
+input_audio = args.input
+output_audio = args.output
+
+if models_dir:
+    ckpt = torch.load(models_dir, map_location=device)
+    generator.load_state_dict(ckpt["G"])
+else:
+    print(
+        "Generator model (--model) isn't specified. Do you have the trained model? If not, you need to train it OR acquire it from somewhere (DON'T ASK ME, YET!)"
+    )
+
+
+def start():
+    # To Mono!
+    decoder = torchcodec.decoders.AudioDecoder(input_audio)
+
+    decoded_samples = decoder.get_all_samples()
+    audio = decoded_samples.data
+    original_sample_rate = decoded_samples.sample_rate
+
+    audio = AudioUtils.stereo_tensor_to_mono(audio)
+    audio = AudioUtils.normalize(audio)
+
+    resample_transform = torchaudio.transforms.Resample(
+        original_sample_rate, args.sample_rate
+    )
+
+    audio = resample_transform(audio)
+
+    splitted_audio = AudioUtils.split_audio(audio, clip_length)
+    splitted_audio_on_device = [t.to(device) for t in splitted_audio]
+    processed_audio = []
+
+    for clip in tqdm.tqdm(splitted_audio_on_device, desc="Processing..."):
+        processed_audio.append(generator(clip))
+
+    reconstructed_audio = AudioUtils.reconstruct_audio(processed_audio)
+    print(f"Saving {output_audio}!")
+    torchaudio.save_with_torchcodec(
+        uri=output_audio,
+        src=reconstructed_audio,
+        sample_rate=args.sample_rate,
+        channels_first=True,
+        compression=args.bitrate,
+    )
+
+
+start()

data.py

@@ -1,50 +1,79 @@
-from torch.utils.data import Dataset
-import torch.nn.functional as F
-import torchaudio
 import os
 import random
 
+import torchaudio
+import torchcodec.decoders as decoders
+import tqdm
+from torch.utils.data import Dataset
+
+import AudioUtils
+
 
 class AudioDataset(Dataset):
-    audio_sample_rates = [8000, 11025, 16000, 22050]
+    audio_sample_rates = [11025]
 
-    def __init__(self, input_dir, target_duration=None, padding_mode='constant', padding_value=0.0):
-        self.input_files = [os.path.join(input_dir, f) for f in os.listdir(input_dir) if f.endswith('.wav')]
-        self.target_duration = target_duration  # Duration in seconds or None if not set
-        self.padding_mode = padding_mode
-        self.padding_value = padding_value
+    def __init__(self, input_dir, clip_length: int = 8000, normalize: bool = True):
+        self.clip_length = clip_length
+        self.normalize = normalize
+
+        input_files = [
+            os.path.join(input_dir, f)
+            for f in os.listdir(input_dir)
+            if os.path.isfile(os.path.join(input_dir, f))
+            and f.lower().endswith((".wav", ".mp3", ".flac"))
+        ]
+
+        data = []
+        for audio_clip in tqdm.tqdm(
+            input_files, desc=f"Processing {len(input_files)} audio file(s)"
+        ):
+            decoder = decoders.AudioDecoder(audio_clip)
+            decoded_samples = decoder.get_all_samples()
+
+            audio = decoded_samples.data.float()  # ensure float32
+            original_sample_rate = decoded_samples.sample_rate
+
+            audio = AudioUtils.stereo_tensor_to_mono(audio)
+            if normalize:
+                audio = AudioUtils.normalize(audio)
+
+            mangled_sample_rate = random.choice(self.audio_sample_rates)
+            resample_transform_low = torchaudio.transforms.Resample(
+                original_sample_rate, mangled_sample_rate
+            )
+            resample_transform_high = torchaudio.transforms.Resample(
+                mangled_sample_rate, original_sample_rate
+            )
+
+            low_audio = resample_transform_high(resample_transform_low(audio))
+
+            splitted_high_quality_audio = AudioUtils.split_audio(audio, clip_length)
+            splitted_low_quality_audio = AudioUtils.split_audio(low_audio, clip_length)
+
+            if not splitted_high_quality_audio or not splitted_low_quality_audio:
+                continue  # skip empty or invalid clips
+
+            splitted_high_quality_audio[-1] = AudioUtils.pad_tensor(
+                splitted_high_quality_audio[-1], clip_length
+            )
+            splitted_low_quality_audio[-1] = AudioUtils.pad_tensor(
+                splitted_low_quality_audio[-1], clip_length
+            )
+
+            for high_quality_data, low_quality_data in zip(
+                splitted_high_quality_audio, splitted_low_quality_audio
+            ):
+                data.append(
+                    (
+                        (high_quality_data, low_quality_data),
+                        (original_sample_rate, mangled_sample_rate),
+                    )
+                )
+
+        self.audio_data = data
 
     def __len__(self):
-        return len(self.input_files)
-
+        return len(self.audio_data)
 
     def __getitem__(self, idx):
-        high_quality_wav, sr_original = torchaudio.load(self.input_files[idx], normalize=True)
-
-        sample_rate = random.choice(self.audio_sample_rates)
-        resample_transform = torchaudio.transforms.Resample(sr_original, sample_rate)
-        low_quality_wav = resample_transform(high_quality_wav)
-        low_quality_wav = low_quality_wav
-
-        # Calculate target length based on desired duration and 16000 Hz
-        if self.target_duration is not None:
-            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_wav = self.stretch_tensor(high_quality_wav, target_length)
-        low_quality_wav = self.stretch_tensor(low_quality_wav, target_length)
-
-
-        return low_quality_wav, high_quality_wav
-
-    def stretch_tensor(self, tensor, target_length):
-        current_length = tensor.size(1)
-        scale_factor = target_length / current_length
-
-        # Resample the tensor using linear interpolation
-        tensor = F.interpolate(tensor.unsqueeze(0), scale_factor=scale_factor, mode='linear', align_corners=False).squeeze(0)
-
-        return tensor
+        return self.audio_data[idx]

discriminator.py

@@ -1,24 +1,75 @@
 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,
+    use_instance_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)
+
+    layers = [conv_layer]
+    layers.append(nn.LeakyReLU(0.2, inplace=True))
+
+    if use_instance_norm:
+        layers.append(nn.InstanceNorm1d(out_channels))
+
+    return nn.Sequential(*layers)
+
+
+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(inplace=True),
+            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=32):
         super(SISUDiscriminator, self).__init__()
         self.model = nn.Sequential(
-            nn.Conv1d(2, 128, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
-            nn.Conv1d(128, 256, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
-            nn.Conv1d(256, 128, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
-            nn.Conv1d(128, 64, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
-            nn.Conv1d(64, 1, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
+            discriminator_block(1, layers, kernel_size=7, stride=1),
+            discriminator_block(layers, layers * 2, kernel_size=5, stride=2),
+            discriminator_block(layers * 2, layers * 4, kernel_size=5, dilation=2),
+            AttentionBlock(layers * 4),
+            discriminator_block(layers * 4, layers * 8, kernel_size=5, dilation=4),
+            discriminator_block(layers * 8, layers * 2, kernel_size=5, stride=2),
+            discriminator_block(
+                layers * 2,
+                1,
+                spectral_norm=False,
+                use_instance_norm=False,
+            ),
         )
-        self.global_avg_pool = nn.AdaptiveAvgPool1d(1)  # Output size (1,)
+
+        self.global_avg_pool = nn.AdaptiveAvgPool1d(1)
 
     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(x.size(0), -1)
         return x

generator.py

@@ -1,23 +1,81 @@
+import torch
 import torch.nn as nn
 
-class SISUGenerator(nn.Module):
-    def __init__(self, upscale_scale=1): # No noise_dim parameter
-        super(SISUGenerator, self).__init__()
-        self.model = nn.Sequential(
-            nn.Conv1d(2, 128, kernel_size=3, padding=1),
-            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'),
+def conv_block(in_channels, out_channels, kernel_size=3, dilation=1):
+    return nn.Sequential(
+        nn.Conv1d(
+            in_channels,
+            out_channels,
+            kernel_size=kernel_size,
+            dilation=dilation,
+            padding=(kernel_size // 2) * dilation,
+        ),
+        nn.InstanceNorm1d(out_channels),
+        nn.PReLU(),
+    )
 
-            nn.Conv1d(256, 128, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
-            nn.Conv1d(128, 64, kernel_size=3, padding=1),
-            nn.LeakyReLU(0.2, inplace=True),
-            nn.Conv1d(64, 2, kernel_size=3, padding=1),
-            nn.Tanh()
+
+class AttentionBlock(nn.Module):
+    """
+    Simple Channel Attention Block. Learns to weight channels based on their importance.
+    """
+
+    def __init__(self, channels):
+        super(AttentionBlock, self).__init__()
+        self.attention = nn.Sequential(
+            nn.Conv1d(channels, channels // 4, kernel_size=1),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(channels // 4, channels, kernel_size=1),
+            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, channels=16, num_rirb=4, alpha=1):
+        super(SISUGenerator, self).__init__()
+        self.alpha = alpha
+
+        self.conv1 = nn.Sequential(
+            nn.Conv1d(1, channels, kernel_size=7, padding=3),
+            nn.InstanceNorm1d(channels),
+            nn.PReLU(),
+        )
+
+        self.rir_blocks = nn.Sequential(
+            *[ResidualInResidualBlock(channels) for _ in range(num_rirb)]
+        )
+
+        self.final_layer = nn.Sequential(
+            nn.Conv1d(channels, 1, kernel_size=3, padding=1), nn.Tanh()
+        )
+
+    def forward(self, x):
+        residual_input = x
+        x = self.conv1(x)
+        x_rirb_out = self.rir_blocks(x)
+        learned_residual = self.final_layer(x_rirb_out)
+        output = residual_input + self.alpha * learned_residual
+
+        return torch.tanh(output)

requirements.txt

@@ -1,12 +0,0 @@
-filelock>=3.16.1
-fsspec>=2024.10.0
-Jinja2>=3.1.4
-MarkupSafe>=2.1.5
-mpmath>=1.3.0
-networkx>=3.4.2
-numpy>=2.1.2
-pillow>=11.0.0
-setuptools>=70.2.0
-sympy>=1.13.1
-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

@@ -1,135 +1,245 @@
+import argparse
+import os
+
 import torch
 import torch.nn as nn
 import torch.optim as optim
-
-import torch.nn.functional as F
-import torchaudio
 import tqdm
-
-from torch.utils.data import random_split
-from torch.utils.data import DataLoader
+from accelerate import Accelerator
+from torch.utils.data import DataLoader, DistributedSampler
 
 from data import AudioDataset
-from generator import SISUGenerator
 from discriminator import SISUDiscriminator
+from generator import SISUGenerator
+from utils.TrainingTools import discriminator_train, generator_train
 
-# Check for CUDA availability
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-print(f"Using device: {device}")
+# ---------------------------
+# Argument parsing
+# ---------------------------
+parser = argparse.ArgumentParser(description="Training script (safer defaults)")
+parser.add_argument("--resume", action="store_true", help="Resume training")
+parser.add_argument(
+    "--epochs", type=int, default=5000, help="Number of training epochs"
+)
+parser.add_argument("--batch_size", type=int, default=8, help="Batch size")
+parser.add_argument("--num_workers", type=int, default=2, help="DataLoader num_workers")
+parser.add_argument("--debug", action="store_true", help="Print debug logs")
+parser.add_argument(
+    "--no_pin_memory", action="store_true", help="Disable pin_memory even on CUDA"
+)
+args = parser.parse_args()
 
-# Initialize dataset and dataloader
-dataset_dir = './dataset/good'
-dataset = AudioDataset(dataset_dir, target_duration=2.0)
+# ---------------------------
+# Init accelerator
+# ---------------------------
 
-dataset_size = len(dataset)
-train_size = int(dataset_size * .9)
-val_size = int(dataset_size-train_size)
+accelerator = Accelerator(mixed_precision="bf16")
 
-train_dataset, val_dataset = random_split(dataset, [train_size, val_size])
-
-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
+# ---------------------------
+# Models
+# ---------------------------
 generator = SISUGenerator()
 discriminator = SISUDiscriminator()
 
-generator = generator.to(device)
-discriminator = discriminator.to(device)
+accelerator.print("🔨 | Compiling models...")
 
-# Loss
-criterion_g = nn.L1Loss()
-criterion_d = nn.BCEWithLogitsLoss()
+generator = torch.compile(generator)
+discriminator = torch.compile(discriminator)
 
-# Optimizers
-optimizer_g = optim.Adam(generator.parameters(), lr=0.0001, betas=(0.5, 0.999))
-optimizer_d = optim.Adam(discriminator.parameters(), lr=0.0001, betas=(0.5, 0.999))
+accelerator.print("✅ | Compiling done!")
 
-# Scheduler
-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)
+# ---------------------------
+# Dataset / DataLoader
+# ---------------------------
+accelerator.print("📊 | Fetching dataset...")
+dataset = AudioDataset("./dataset")
 
-def snr(y_true, y_pred):
-    noise = y_true - y_pred
-    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
+sampler = DistributedSampler(dataset) if accelerator.num_processes > 1 else None
+pin_memory = torch.cuda.is_available() and not args.no_pin_memory
 
-def discriminator_train(discriminator, optimizer, criterion, generator, real_labels, fake_labels, high_quality, low_quality):
-    optimizer.zero_grad()
+train_loader = DataLoader(
+    dataset,
+    sampler=sampler,
+    batch_size=args.batch_size,
+    shuffle=(sampler is None),
+    num_workers=args.num_workers,
+    pin_memory=pin_memory,
+    persistent_workers=pin_memory,
+)
 
-    discriminator_decision_from_real = discriminator(high_quality)
-    d_loss_real = criterion(discriminator_decision_from_real, real_labels)
+if not train_loader or not train_loader.batch_size or train_loader.batch_size == 0:
+    accelerator.print("🪹 | There is no data to train with! Exiting...")
+    exit()
 
-    generator_output = generator(low_quality)
-    discriminator_decision_from_fake = discriminator(generator_output.detach())
-    d_loss_fake = criterion(discriminator_decision_from_fake, fake_labels)
+loader_batch_size = train_loader.batch_size
 
-    d_loss = (d_loss_real + d_loss_fake) / 2.0
+accelerator.print("✅ | Dataset fetched!")
 
-    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}")
+# ---------------------------
+# Losses / Optimizers / Scalers
+# ---------------------------
 
-def start_training():
+optimizer_g = optim.AdamW(
+    generator.parameters(), lr=0.0003, betas=(0.5, 0.999), weight_decay=0.0001
+)
+optimizer_d = optim.AdamW(
+    discriminator.parameters(), lr=0.0003, betas=(0.5, 0.999), weight_decay=0.0001
+)
 
-    # Training loop
-    # discriminator_epochs = 1000
-    generator_epochs = 500
-    for generator_epoch in range(generator_epochs):
-        low_quality_audio = torch.empty((1))
-        high_quality_audio = torch.empty((1))
-        ai_enhanced_audio = torch.empty((1))
+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
+)
 
-        # Training
-        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)
+criterion_g = nn.BCEWithLogitsLoss()
+criterion_d = nn.MSELoss()
 
+# ---------------------------
+# Prepare accelerator
+# ---------------------------
+
+generator, discriminator, optimizer_g, optimizer_d, train_loader = accelerator.prepare(
+    generator, discriminator, optimizer_g, optimizer_d, train_loader
+)
+
+# ---------------------------
+# Checkpoint helpers
+# ---------------------------
+models_dir = "./models"
+os.makedirs(models_dir, exist_ok=True)
+
+
+def save_ckpt(path, epoch):
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        accelerator.save(
+            {
+                "epoch": epoch,
+                "G": accelerator.unwrap_model(generator).state_dict(),
+                "D": accelerator.unwrap_model(discriminator).state_dict(),
+                "optG": optimizer_g.state_dict(),
+                "optD": optimizer_d.state_dict(),
+                "schedG": scheduler_g.state_dict(),
+                "schedD": scheduler_d.state_dict(),
+            },
+            path,
+        )
+
+
+start_epoch = 0
+if args.resume:
+    ckpt_path = os.path.join(models_dir, "last.pt")
+    ckpt = torch.load(ckpt_path)
+
+    accelerator.unwrap_model(generator).load_state_dict(ckpt["G"])
+    accelerator.unwrap_model(discriminator).load_state_dict(ckpt["D"])
+    optimizer_g.load_state_dict(ckpt["optG"])
+    optimizer_d.load_state_dict(ckpt["optD"])
+    scheduler_g.load_state_dict(ckpt["schedG"])
+    scheduler_d.load_state_dict(ckpt["schedD"])
+
+    start_epoch = ckpt.get("epoch", 1)
+    accelerator.print(f"🔁 | Resumed from epoch {start_epoch}!")
+
+real_buf = torch.full(
+    (loader_batch_size, 1), 1, device=accelerator.device, dtype=torch.float32
+)
+fake_buf = torch.zeros(
+    (loader_batch_size, 1), device=accelerator.device, dtype=torch.float32
+)
+
+accelerator.print("🏋️ | Started training...")
+
+try:
+    for epoch in range(start_epoch, args.epochs):
+        generator.train()
+        discriminator.train()
+
+        running_d, running_g, steps = 0.0, 0.0, 0
+
+        for i, (
+            (high_quality, low_quality),
+            (high_sample_rate, low_sample_rate),
+        ) in enumerate(tqdm.tqdm(train_loader, desc=f"Epoch {epoch}")):
             batch_size = high_quality.size(0)
-            real_labels = torch.ones(batch_size, 1).to(device)
-            fake_labels = torch.zeros(batch_size, 1).to(device)
 
-            # Train Discriminator
-            discriminator.train()
+            real_labels = real_buf[:batch_size].to(accelerator.device)
+            fake_labels = fake_buf[:batch_size].to(accelerator.device)
 
-            for _ in range(3):
-                discriminator_train(discriminator, optimizer_d, criterion_d, generator, real_labels, fake_labels, high_quality, low_quality)
+            # --- Discriminator ---
+            optimizer_d.zero_grad(set_to_none=True)
+            with accelerator.autocast():
+                d_loss = discriminator_train(
+                    high_quality,
+                    low_quality,
+                    real_labels,
+                    fake_labels,
+                    discriminator,
+                    generator,
+                    criterion_d,
+                )
 
-            # Train Generator
-            generator.train()
-            optimizer_g.zero_grad()
+            accelerator.backward(d_loss)
+            torch.nn.utils.clip_grad_norm_(discriminator.parameters(), 1)
+            optimizer_d.step()
 
-            # Generator loss: how well fake data fools the discriminator
-            generator_output = generator(low_quality)
-            discriminator_decision = discriminator(generator_output)  # No detach here
-            g_loss = criterion_g(discriminator_decision, real_labels)  # Train generator to produce real-like outputs
+            # --- Generator ---
+            optimizer_g.zero_grad(set_to_none=True)
+            with accelerator.autocast():
+                g_total, g_adv = generator_train(
+                    low_quality,
+                    high_quality,
+                    real_labels,
+                    generator,
+                    discriminator,
+                    criterion_d,
+                )
 
-            g_loss.backward()
+            accelerator.backward(g_total)
+            torch.nn.utils.clip_grad_norm_(generator.parameters(), 1)
             optimizer_g.step()
 
-            low_quality_audio = low_quality
-            high_quality_audio = high_quality
-            ai_enhanced_audio = generator_output
+            d_val = accelerator.gather(d_loss.detach()).mean()
+            g_val = accelerator.gather(g_total.detach()).mean()
 
-        metric = snr(high_quality_audio, ai_enhanced_audio)
-        print(f"Generator metric {metric}!")
-        scheduler_g.step(metric)
+            if torch.isfinite(d_val):
+                running_d += d_val.item()
+            else:
+                accelerator.print(
+                    f"🫥 | NaN in discriminator loss at step {i}, skipping update."
+                )
 
-        if generator_epoch % 10 == 0:
-            print(f"Saved epoch {generator_epoch}!")
-            torchaudio.save(f"./output/epoch-{generator_epoch}-audio-crap.wav", low_quality_audio[0].cpu(), 44100)
-            torchaudio.save(f"./output/epoch-{generator_epoch}-audio-ai.wav", ai_enhanced_audio[0].cpu(), 44100)
-            torchaudio.save(f"./output/epoch-{generator_epoch}-audio-orig.wav", high_quality_audio[0].cpu(), 44100)
+            if torch.isfinite(g_val):
+                running_g += g_val.item()
+            else:
+                accelerator.print(
+                    f"🫥 | NaN in generator loss at step {i}, skipping update."
+                )
 
-        if generator_epoch % 50 == 0:
-            torch.save(discriminator.state_dict(), "discriminator.pt")
-            torch.save(generator.state_dict(), "generator.pt")
+            steps += 1
 
-    torch.save(discriminator.state_dict(), "discriminator.pt")
-    torch.save(generator.state_dict(), "generator.pt")
-    print("Training complete!")
+        # epoch averages & schedulers
+        if steps == 0:
+            accelerator.print("🪹 | No steps in epoch (empty dataloader?). Exiting.")
+            break
 
-start_training()
+        mean_d = running_d / steps
+        mean_g = running_g / steps
+
+        scheduler_d.step(mean_d)
+        scheduler_g.step(mean_g)
+
+        save_ckpt(os.path.join(models_dir, "last.pt"), epoch)
+        accelerator.print(f"🤝 | Epoch {epoch} done | D {mean_d:.4f} | G {mean_g:.4f}")
+
+except Exception:
+    try:
+        save_ckpt(os.path.join(models_dir, "crash_last.pt"), epoch)
+        accelerator.print(f"💾 | Saved crash checkpoint for epoch {epoch}")
+    except Exception as e:
+        accelerator.print("😬 | Failed saving crash checkpoint:", e)
+    raise
+
+accelerator.print("🏁 | Training finished.")

utils/MultiResolutionSTFTLoss.py

@@ -0,0 +1,87 @@
+from typing import Dict, List
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torchaudio.transforms as T
+
+
+class MultiResolutionSTFTLoss(nn.Module):
+    """
+    Multi-resolution STFT loss.
+    Combines spectral convergence loss and log-magnitude loss
+    across multiple STFT resolutions.
+    """
+
+    def __init__(
+        self,
+        fft_sizes: List[int] = [1024, 2048, 512],
+        hop_sizes: List[int] = [120, 240, 50],
+        win_lengths: List[int] = [600, 1200, 240],
+        eps: float = 1e-7,
+    ):
+        super().__init__()
+
+        self.eps = eps
+        self.n_resolutions = len(fft_sizes)
+
+        self.stft_transforms = nn.ModuleList()
+        for n_fft, hop_len, win_len in zip(fft_sizes, hop_sizes, win_lengths):
+            window = torch.hann_window(win_len)
+            stft = T.Spectrogram(
+                n_fft=n_fft,
+                hop_length=hop_len,
+                win_length=win_len,
+                window_fn=lambda _: window,
+                power=None,  # Keep complex output
+                center=True,
+                pad_mode="reflect",
+                normalized=False,
+            )
+            self.stft_transforms.append(stft)
+
+    def forward(
+        self, y_true: torch.Tensor, y_pred: torch.Tensor
+    ) -> Dict[str, torch.Tensor]:
+        """
+        Args:
+            y_true: (B, T) or (B, 1, T) waveform
+            y_pred: (B, T) or (B, 1, T) waveform
+        """
+        # Ensure correct shape (B, T)
+        if y_true.dim() == 3 and y_true.size(1) == 1:
+            y_true = y_true.squeeze(1)
+        if y_pred.dim() == 3 and y_pred.size(1) == 1:
+            y_pred = y_pred.squeeze(1)
+
+        sc_loss = 0.0
+        mag_loss = 0.0
+
+        for stft in self.stft_transforms:
+            stft = stft.to(y_pred.device)
+
+            # Complex STFTs: (B, F, T, 2)
+            stft_true = stft(y_true)
+            stft_pred = stft(y_pred)
+
+            # Magnitudes
+            stft_mag_true = torch.abs(stft_true)
+            stft_mag_pred = torch.abs(stft_pred)
+
+            # --- Spectral Convergence Loss ---
+            norm_true = torch.linalg.norm(stft_mag_true, dim=(-2, -1))
+            norm_diff = torch.linalg.norm(stft_mag_true - stft_mag_pred, dim=(-2, -1))
+            sc_loss += torch.mean(norm_diff / (norm_true + self.eps))
+
+            # --- Log STFT Magnitude Loss ---
+            mag_loss += F.l1_loss(
+                torch.log(stft_mag_pred + self.eps),
+                torch.log(stft_mag_true + self.eps),
+            )
+
+        # Average across resolutions
+        sc_loss /= self.n_resolutions
+        mag_loss /= self.n_resolutions
+        total_loss = sc_loss + mag_loss
+
+        return {"total": total_loss, "sc": sc_loss, "mag": mag_loss}

utils/TrainingTools.py

@@ -0,0 +1,60 @@
+import torch
+
+# In case if needed again...
+# from utils.MultiResolutionSTFTLoss import MultiResolutionSTFTLoss
+#
+# stft_loss_fn = MultiResolutionSTFTLoss(
+#     fft_sizes=[1024, 2048, 512], hop_sizes=[120, 240, 50], win_lengths=[600, 1200, 240]
+# )
+
+
+def signal_mae(input_one: torch.Tensor, input_two: torch.Tensor) -> torch.Tensor:
+    absolute_difference = torch.abs(input_one - input_two)
+    return torch.mean(absolute_difference)
+
+
+def discriminator_train(
+    high_quality,
+    low_quality,
+    high_labels,
+    low_labels,
+    discriminator,
+    generator,
+    criterion,
+):
+    decision_high = discriminator(high_quality)
+    d_loss_high = criterion(decision_high, high_labels)
+    # print(f"Is this real?: {discriminator_decision_from_real} | {d_loss_real}")
+
+    decision_low = discriminator(low_quality)
+    d_loss_low = criterion(decision_low, low_labels)
+    # print(f"Is this real?: {discriminator_decision_from_fake} | {d_loss_fake}")
+
+    with torch.no_grad():
+        generator_quality = generator(low_quality)
+    decision_gen = discriminator(generator_quality)
+    d_loss_gen = criterion(decision_gen, low_labels)
+
+    noise = torch.rand_like(high_quality) * 0.08
+    decision_noise = discriminator(high_quality + noise)
+    d_loss_noise = criterion(decision_noise, low_labels)
+
+    d_loss = (d_loss_high + d_loss_low + d_loss_gen + d_loss_noise) / 4.0
+
+    return d_loss
+
+
+def generator_train(
+    low_quality, high_quality, real_labels, generator, discriminator, adv_criterion
+):
+    generator_output = generator(low_quality)
+
+    discriminator_decision = discriminator(generator_output)
+    adversarial_loss = adv_criterion(discriminator_decision, real_labels)
+
+    # Signal similarity
+    similarity_loss = signal_mae(generator_output, high_quality)
+
+    combined_loss = adversarial_loss + (similarity_loss * 100)
+
+    return combined_loss, adversarial_loss