nullableeth/rvc-trainer repository overview

# Prepare your audio (10-15 minutes of clean speech recommended)
mkdir -p training/input
cp your_voice.wav training/input/

# Start training container
docker run -it --rm --gpus all \
  --shm-size=8g \
  -v $(pwd)/training:/app/training \
  nullableeth/rvc-trainer:latest \
  bash

# Inside container - preprocess audio
cd /app/rvc
python3 infer/modules/train/preprocess.py \
  /app/training/input \
  40000 \
  4 \
  /app/training/my_voice \
  False \
  1.0

# Extract features (run all 4 parts)
for part in 0 1 2 3; do
  python3 infer/modules/train/extract_feature_print.py \
    cuda:0 4 $part /app/training/my_voice v2 False
done

# Extract pitch
python3 infer/modules/train/extract/extract_f0_print.py \
  /app/training/my_voice \
  4 \
  harvest

# Generate filelist
python3 << 'PYEOF'
import os
exp_dir = '/app/training/my_voice'
feature_files = sorted([f.replace('.npy', '') for f in os.listdir(f'{exp_dir}/3_feature768') if f.endswith('.npy')])
with open(f'{exp_dir}/filelist.txt', 'w') as f:
    for basename in feature_files:
        gt_wav = f"{exp_dir}/0_gt_wavs/{basename}.wav"
        phone = f"{exp_dir}/3_feature768/{basename}.npy"
        f0 = f"{exp_dir}/2a_f0/{basename}.wav.npy"
        f0nsf = f"{exp_dir}/2b-f0nsf/{basename}.wav.npy"
        if all(os.path.exists(p) for p in [gt_wav, phone, f0, f0nsf]):
            f.write(f"{gt_wav}|{phone}|{f0}|{f0nsf}|0\n")
PYEOF

# Create config (see full example below)
# Then train...
python3 infer/modules/train/train.py \
  -e /app/training/my_voice \
  -sr 40k \
  -f0 1 \
  -bs 2 \
  -g 0 \
  -te 200 \
  -se 50 \
  -pg assets/pretrained_v2/f0G40k.pth \
  -pd assets/pretrained_v2/f0D40k.pth \
  -l 0 \
  -c 0 \
  -sw 0 \
  -v v2

{
  "train": {
    "log_interval": 200,
    "seed": 1234,
    "epochs": 200,
    "learning_rate": 1e-4,
    "betas": [0.8, 0.99],
    "eps": 1e-9,
    "batch_size": 2,
    "fp16_run": false,
    "lr_decay": 0.999875,
    "segment_size": 12800,
    "init_lr_ratio": 1,
    "warmup_epochs": 0,
    "c_mel": 45,
    "c_kl": 1.0
  },
  "data": {
    "max_wav_value": 32768.0,
    "sampling_rate": 40000,
    "filter_length": 2048,
    "hop_length": 400,
    "win_length": 2048,
    "n_mel_channels": 125,
    "mel_fmin": 0.0,
    "mel_fmax": null
  },
  "model": {
    "inter_channels": 192,
    "hidden_channels": 192,
    "filter_channels": 768,
    "n_heads": 2,
    "n_layers": 6,
    "kernel_size": 3,
    "p_dropout": 0,
    "resblock": "1",
    "resblock_kernel_sizes": [3,7,11],
    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
    "upsample_rates": [10,10,2,2],
    "upsample_initial_channel": 512,
    "upsample_kernel_sizes": [16,16,4,4],
    "use_spectral_norm": false,
    "gin_channels": 256,
    "spk_embed_dim": 109
  },
  "version": "v2",
  "sr": "40k",
  "if_f0": 1,
  "spk": {
    "my_voice": 0
  }
}

# Change -l 0 to -l 1 and increase -te
python3 infer/modules/train/train.py \
  -e /app/training/my_voice \
  -sr 40k \
  -f0 1 \
  -bs 2 \
  -g 0 \
  -te 500 \
  -se 50 \
  -pg assets/pretrained_v2/f0G40k.pth \
  -pd assets/pretrained_v2/f0D40k.pth \
  -l 1 \
  -c 0 \
  -sw 0 \
  -v v2