Automatic calibration of command-detection thresholds
US8990079B1 · kind B1 · utility
Assignee
Inventor
Key dates
| Filing date | Sep 17, 2014 |
| Grant date | Mar 24, 2015 |
| Priority date | — |
| Expiry date | Sep 17, 2034 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG10L2025/786
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
When a voice-activated device or application is first started, the signal levels corresponding to spoken commands are initially unknown, making it difficult to set detection thresholds. The inventive method provides an initial command-detection threshold based on the noise level alone. The first command is detected using this initial threshold. Then the threshold is revised according to the first command sound, and a second command is detected using the revised threshold. After detecting each command, the detection threshold is further refined according to the current noise and command sounds. Methods are also disclosed for optimizing the thresholds, adjusting parameters according to sound, and detecting voiced and unvoiced sounds separately. These capabilities enable many emerging voice-activated products and applications.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.