Not include a significant amount of music (background or foreground) or noise. If you have longer files that contain sections of music, consider splitting the files to remove these sections.
Contain speech that is relatively clear and well articulated. Speech that is difficult to understand for a human ear is similarly challenging for the automated system.
Not include speakers talking over each other; this can cause problems if it occurs frequently.
Be a close match in quality and nature to the audio that you ultimately expect to process.
Note: HPE IDOL Speech Server can analyze audio files and produce information about audio quality, such as SNR, clipping, presence of music, and noise. For more information about analyzing audio, see Preprocess Audio.