Evaluate "input_confidence" vs "fingerprinted_confidence"
Opened this issue · 0 comments
denis-stepanov commented
To decide on a hit, AdVent uses "fingerprinted_confidence" result returned by Dejavu. Its calculation depends inversely on the length of the fingerprinted track, i.e. longer jingles would get lower confidence, and one would need to increase listening interval to compensate. It probably makes sense to evaluate "input_confidence" instead, which presumably is calculated as a ratio of fingerprints matched to fingerprints submitted (to be confirmed). If true, it might be more suitable for AdVent purpose.