Our Technology

Our technology uses advanced machine learning to detect sensory triggers in audio content, providing users with tailored warnings. We start by transforming audio into feature-rich embeddings using cutting-edge audio processing models like OpenL3 or Wav2Vec2. At the same time, potential trigger descriptions (e.g., "clap," "scream," "horn") are converted into numerical representations using state-of-the-art text embedding models such as Sentence-BERT. By comparing these representations through cosine similarity, we calculate how closely the audio matches any trigger label. If the similarity exceeds a set threshold, our system flags the segment and warns the user. This approach ensures precision and personalization, helping individuals with sensory sensitivities enjoy their content with confidence.