Spectrograms and Sentiment Analysis
Spectrograms are one of the most commonly used ways to analyze speech signals. However, they are generally not used as evidence in trials for forensic purposes. They may be used by an expert witness, such as a linguist/phonetician, but this person bears witness as an expert and spectrograms are not presented as evidence. The reason why they do not use spectrograms as evidence is that they are somewhat technical and can be used to convince jurors, one way or another, who might not understand the technical details. (As an aside, technical details may also convince a defendant or plaintiff: My forensic linguistics professor, Prof. Disner, was an expert witness for the famous lawyer Johnie Cochran and he asked her to bring all her technical recording and analysis equipment to the trial, which the the opposing party saw and became intimidated and they promptly confessed/settled). The word “spectrogram” came to be a bad word in court cases because they were overused (misused) in this way, probably law enforcement over using it to get convictions in cases where recordings were used as evidence.
I think that sentiment analysis could be a similar phenomenon, if it is overused or used naively. It might be possible to overuse sentiment analysis and lead people to dismiss it in a similar way. I think that one thing that would be good to remedy this possible risk is to release benchmark data sets so that different algorithms can be rigorously tested.
Originally published on abewrites.
clustrmaps.html