- Media recognition
- Media data analysis
- Anomaly detection
Uncommon Sound Event Detection Technology
This technology monitors sounds and detects "that sound" you want to focus on.
- Based on a small amount of sample data of the target sound event ("that sound"), it detects the sound event from input acoustic signals (few-shot sound event detection).
- If the target sound event is a well-known concept (e.g., "scream", "explosion", "music", etc.), it can be detected by text input without sample data (zero-shot sound event detection).
- If no sample data is available, this technology supports the search for samples of the target sound event. Short-term (several minutes) data of infrequent sound events can be extracted from long-term (several hours to days) on-site recordings (uncommon sound detection). This enables efficient searching for samples of the target event within a short period.
Applications
- Monitoring and visualization of maintenance operations
- Acoustic monitoring of plants and public facilities
Benchmarks, strengths, and track record
- Target sound events can be detected by specifying a small amount of sample sound or a text description.
- Even if the target acoustic event occurs infrequently and recording samples is difficult, it can be easily found from the sound you have collected.
Inquiries
Inquiries to Toshiba Corporate Laboratory (Komukai region)
Please include the title “Toshiba AI Technology Catalog: Uncommon Sound Event Detection Technology” or the URL in the inquiry text.
Please note that because this technology is currently the subject of R&D activities, immediate responses to inquiries may not be possible.
