Sarcasm can be tricky for even humans to pick up — let alone a computer.
That’s why researchers at the University of Groningen’s Speech Technology Lab decided to build an AI sarcasm detector that can pick up on tone of voice and convey those emotions through emojis embedded in transcribed text.
One of the researchers who worked on the project, Xiyuan Gao, presented the work on Thursday as part of a joint meeting held by the Acoustical Society of America and the Canadian Acoustical Association at the Shaw Center in Ottawa.
Usually, sentiment analysis just “focuses on text,” according to Gao.
The new approach goes deeper into the way people say things, not just what they say, which could help fields like AI-assisted health care. The findings of the study could also mean better AI virtual assistants that can pick up on tone.
Related: These ‘Expressive Avatar’ Deepfakes From a Billion-Dollar AI Startup Look Scary Real
The study took a multilayered approach to sarcasm, evaluating both what they could hear and what the speaker said on paper.
The researchers first evaluated audio recordings based on pitch, speaking rate, and other factors to figure out the emotions underneath each word.
They then transcribed the audio recordings into text and labeled each text segment with emojis that reflected the emotional intent behind the speech.
“Our approach leverages the combined strengths of auditory and textual information along with emoticons for a comprehensive analysis,” Gao stated in a press release.
Related: Employers Say They Want to Hire Candidates With AI Skills, But Employees Are Still Sneaking AI Tool Use in the Office
Looking ahead, the researchers want their algorithm to be able to pick up on more sarcastic expressions and gestures.
“In addition, we would like to include more languages,” Gao said.
AI voice cloning and generation has been top of mind recently as OpenAI, Google and other tech companies release cutting-edge AI models with more emotive voices than ever.
OpenAI showcased Voice Engine last month, but held back on releasing the text-to-speech realistic voice generator because of “the potential for synthetic voice misuse.”
Related: OpenAI Is Holding Back the Release of Its New AI Voice Generator — Here’s Why
Other projects presented at the acoustic conference include spiderwebs in microphones and ways to reduce noise in social settings.
