from pathlib import Path
import pandas as pd
import torch
import torchaudio
from .superb_sid import SuperbSID
torchaudio.set_audio_backend("sox_io")
[docs]class CommonExample(SuperbSID):
[docs] def default_config(self) -> dict:
config = super().default_config()
config["prepare_data"] = {}
config["train"] = dict(
total_steps=10,
log_step=1,
eval_step=5,
save_step=5,
gradient_clipping=1.0,
gradient_accumulate=1,
valid_metric="accuracy",
valid_higher_better=True,
auto_resume=True,
)
return config
[docs] def prepare_data(
self,
prepare_data: dict,
target_dir: str,
cache_dir: str,
get_path_only: bool = False,
):
target_dir: Path = Path(target_dir)
wavs = [torch.randn(1, 16000 * 2) for i in range(5)]
wav_paths = []
for idx, wav in enumerate(wavs):
wav_path = str(Path(target_dir) / f"{idx}.wav")
torchaudio.save(wav_path, wav, sample_rate=16000)
wav_paths.append(wav_path)
ids = [Path(path).stem for path in wav_paths]
labels = ["a", "a", "b", "c", "d"]
df = pd.DataFrame({"id": ids, "wav_path": wav_paths, "label": labels})
train_csv, valid_csv, test_csv = (
target_dir / "train.csv",
target_dir / "valid.csv",
target_dir / "test.csv",
)
df.iloc[:3].to_csv(train_csv)
df.iloc[3:4].to_csv(valid_csv)
df.iloc[4:].to_csv(test_csv)
return train_csv, valid_csv, [test_csv]