This article announces Amazon Transcribe Toxicity Detection, an ML-powered feature that identifies harmful language in spoken conversations across seven categories.

• Detects toxic content in seven categories: profanity, hate speech, sexual content, insults, violence/threats, graphic language, harassment/abuse
• Uses both audio cues (tone, pitch) and text analysis to identify toxic intent
• Reduces content moderators' review workload by 95% by flagging specific toxic segments
• Enables SLA reduction from 7-15 days to just a few hours
• Available via Amazon Transcribe console, AWS CLI, and Python SDK
• Supports batch processing for US English language only
• Provides confidence scores (0-1) for each toxicity category
• Available in six AWS regions including US East, US West, Asia Pacific, and Europe

Amazon Transcribe Toxicity Detection automates content moderation at scale, helping organizations maintain safe online environments while reducing moderator workload and psychological impact.