train.py

import os
import glob
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
import librosa
import soundfile as sf
import numpy as np

# -----------------------------
# Dataset
# -----------------------------
class AudioDataset(Dataset):
    def __init__(self, dataset_path, sample_rate=44100, duration_sec=2.0):
        self.sample_rate = sample_rate
        self.duration = int(sample_rate * duration_sec)

        self.files = sorted(glob.glob(os.path.join(dataset_path, "**/*.wav"), recursive=True))
        if len(self.files) == 0:
            raise ValueError("No WAV files found in dataset.")

    def __len__(self):
        return len(self.files)

    def __getitem__(self, idx):
        path = self.files[idx]
        audio, sr = sf.read(path)

        # Resample if needed
        if sr != self.sample_rate:
            audio = librosa.resample(audio, orig_sr=sr, target_sr=self.sample_rate)

        # Pad/trim
        if len(audio) < self.duration:
            audio = np.pad(audio, (0, self.duration - len(audio)))
        else:
            audio = audio[:self.duration]

        audio = torch.tensor(audio, dtype=torch.float32)
        return audio.unsqueeze(0)   # (1, samples)


# -----------------------------
# Simple placeholder model
# (Replace with real RVC model)
# -----------------------------
class SimpleVoiceModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.encoder = nn.Sequential(
            nn.Conv1d(1, 32, 3, padding=1),
            nn.ReLU(),
            nn.Conv1d(32, 64, 3, padding=1),
            nn.ReLU(),
        )
        self.decoder = nn.Sequential(
            nn.Conv1d(64, 32, 3, padding=1),
            nn.ReLU(),
            nn.Conv1d(32, 1, 3, padding=1)
        )

    def forward(self, x):
        z = self.encoder(x)
        out = self.decoder(z)
        return out


# -----------------------------
# Training Loop
# -----------------------------
def train(
    dataset_path="dataset",
    batch_size=4,
    epochs=5,
    lr=1e-4,
    save_every=1,
    device="cuda" if torch.cuda.is_available() else "cpu"
):
    print("Loading dataset...")
    ds = AudioDataset(dataset_path)
    dl = DataLoader(ds, batch_size=batch_size, shuffle=True)

    print("Initializing model...")
    model = SimpleVoiceModel().to(device)
    optimizer = optim.Adam(model.parameters(), lr=lr)
    criterion = nn.L1Loss()

    os.makedirs("checkpoints", exist_ok=True)

    print("Training...")
    for epoch in range(1, epochs + 1):
        total_loss = 0

        for batch in dl:
            batch = batch.to(device)

            optimizer.zero_grad()
            out = model(batch)
            loss = criterion(out, batch)   # Autoencoder-style loss
            loss.backward()
            optimizer.step()

            total_loss += loss.item()

        avg_loss = total_loss / len(dl)
        print(f"Epoch {epoch}/{epochs} | Loss: {avg_loss:.6f}")

        if epoch % save_every == 0:
            ckpt_path = f"checkpoints/model_epoch_{epoch}.pt"
            torch.save(model.state_dict(), ckpt_path)
            print(f"Saved checkpoint: {ckpt_path}")

    print("Training complete!")


# -----------------------------
# Entry point (called by app.py)
# -----------------------------
if __name__ == "__main__":
    train(
        dataset_path="dataset",
        batch_size=4,
        epochs=10,
        lr=1e-4,
        save_every=1
    )