Video: Detecting Deception in Speech, Pt. 2: Findings
Learn more about speech technology in academia at the next SpeechTEK conference.
Read the complete transcript of this clip:
Yocheved Levitan: Through studying this corpus and the different features that they extracted, these were some of the findings that they saw. In deceptive speech, there was increased pitch and increased volume, lack of speech planning, detailed speech, whereas with truthful, there was indication of cognitive process and function words that were used. These were useful features for distinguishing between truth and falsehood.
\Then with these results, seeing that there are features that we can use to distinguish between true and false statements, they looked at the different feature sets and relative to the units of speech, and these are some of the results. So on the x-axis are the different feature sets and different combinations, and on the y-axis is the F1 score, which is a measure of precision and recall. So for the interposal units, you see that the performance isn't that great, just above chance.
Looking at longer segments of speech, performance increases. And then there's a big jump when they looked at the question responses, which include even more context. So, and finally question chunks gave the highest performance because it includes the most context and there's the most information. And there we see that it's about 70%, 70 F score. Comparing with the human baseline, we see that this approach significantly outperforms human performance.
Some things to touch on in terms of individual differences. So we mentioned that there was increases in pitch and intensity in deceptive speech, but those differed across gender. Males increased their pitch, while females increased their intensity while lying. For native language speakers, there was a feature for complexity. But since both native language English speakers and Mandarin Chinese speakers were both speaking English, you can't really take the complexity into account because it's harder for people who weren't born speaking English to produce English sentences. And then with neuroticism, that was also another feature for personality that. Yes?
Audience member: Can you explain what neuroticism is, basically?
Yocheved Levitan: It's like neurotic, being paranoid, things like that. So there are five personality factors, openness to experience. It's an acronym OCEAN, and they looked at those, and didn't find many significant results for the personality scores, but for neuroticism, there was a correlation.
Nuance Communications Senior Manager, Commercial Security Strategy Roanne Levitt explains how Nuance's Conversation Print addresses speech spoofing in this clip from her panel an SpeechTEK 2019.
Nuance Communications Senior Manager, Commercial Security Strategy Roanne Levitt and ID R&D VP of Sales John Amein discuss the essential requirements for complying with privacy regulations in this clip from their panel at SpeechTEK 2019.
Pindrop Director of Product Marketing Ben Cunningham discusses how liveness detection, synthetic speech detection, and other deep voice biometric technologies can fight the threat deepfakes pose in this clip from a panel at SpeechTEK 2019.
Interactions LLC NLP Scientist Yocheved Levitan discusses a recent research study on identifying cues of deception in speech, and machine learning-based approaches to classifying deceptive speech in this clip from her presentation at SpeechTEK 2019.