# Assuming a custom dataset class `AudioDataset` for loading WAV files and converting them into spectrograms
| Feature | | MP3 (Avoid) | | :--- | :--- | :--- | | Compression | Lossless (no data removed) | Lossy (data removed to save space) | | Signal Quality | Clean, sharp, predictable | Blurred, softened, unpredictable | | The Risk | None (safe for estim) | MP3 codecs can create random spikes or "clicks" that feel like sharp, unexpected stings. | estim wav files
Most likely, you are looking for either (used in speech recognition) or E-Prime E-Studio files (used in psychology research), or perhaps you meant Stem files (used in music production). # Assuming a custom dataset class `AudioDataset` for