Summary of An Investigation Into Explainable Audio Hate Speech Detection, by Jinmyeong An et al.
An Investigation Into Explainable Audio Hate Speech Detection
by Jinmyeong An, Wonjun Lee, Yejin Jeon, Jungseul Ok, Yunsu Kim, Gary Geunbae Lee
First submitted to arxiv on: 12 Aug 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Sound (cs.SD); Audio and Speech Processing (eess.AS)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This research paper introduces a new task of explainable audio hate speech detection, which involves identifying precise time intervals in verbal content that serve as evidence for hate speech classification. The authors propose two approaches: cascading and End-to-End (E2E). The cascading method converts audio to transcripts, identifies hate speech, and then locates the corresponding audio time frames. In contrast, the E2E approach processes audio utterances directly, pinpointing hate speech within specific time frames. To train these models, a synthetic audio dataset is curated, and actual human speech utterances are used for validation. The results show that the E2E approach outperforms the cascading method in terms of the audio frame Intersection over Union (IoU) metric. Additionally, including frame-level rationales significantly enhances hate speech detection accuracy for the E2E approach. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding hateful words and phrases in spoken language. Right now, most research focuses on written text, but this study looks at verbal content instead. The goal is to identify specific moments in audio recordings that make it clear whether someone is saying something hateful or not. The researchers propose two ways to do this: one method involves converting the audio into text first, and then looking for hate speech, while the other approach processes the audio directly and identifies the problematic parts. To train these methods, they created a fake dataset of audio recordings with hate speech, and then tested them on real human speeches. The results show that one method is better than the other at finding hateful moments in audio. |
Keywords
» Artificial intelligence » Classification
