The thumbs up and down approach to rating an audio submission is not fine grained enough. Need evaluations of sound based on new sets of parameters, so that acoustic models can be created based on creating a sub-corpus that includes all clean speech audio, or or a sub-corpus that includes audio that has noise, but which better reflects the user environment.
Possible parameters to subjectively describe a user submision include:
- Noise characteristics:
- Hiss - low-medium-high (or 1 - 10)
- External noise
- Echo
- non-speech noise
- audio characteristics
- Avg volume - low-medium-high (or 1-10)
- Speech characteristics
- dialect (drop down list, voting system for selection of dialect)
- accent
