Robustness of Statistical Voice Conversion Based on Direct Waveform Modification Against Background Sounds