We have a Guide For Contributors.
Training examples
Please refer to these examples frequently, both when you are seeking source recordings and when you are editing clips to contribute to the database.
First, here are two samples of "real" voices from our preliminary database. The Queen Latifah samples are nearly perfect for VoxDB; the voice is clear and undistorted, and the background noise level is extremely low. The content of the J.K. Rowling samples is fine, but the sound quality is not acceptable for several reasons, including both distortion and background noise, and these samples will need to be replaced. These are both contemporary celebrities, and it should be relatively easy to find far better recordings of J.K. Rowling.
Next, we have prepared some focused examples demonstrating some of the target parameters for VoxDB clips, and some of the specific problems you are likely to encounter.
- Acceptable clip, with high voice level, low background noise, and no distortion.
- Unacceptable clip, recorded too quietly
- Unacceptable clip, recorded too "hot": overmodulated ("clipped" or "distorted")
- Unacceptable clip, with too much background noise (hiss, hum, etc.)
- Unacceptable clip, with human noise in the background
- Unacceptable clip, with muffled sound
- Unacceptable clip, with bad edits resulting in clicks or pops
- Unacceptable clip, with "popped" plosive consonants (most often "p" and "b")
- Unacceptable clip, with bad edits - stopping in the middle of a phrase
- Unacceptable clip, with two voices
- Unacceptable clip, with highly emotional content, shouting, etc.