Browse Prior Art Database

Background Optimization of Speech Recognition Disclosure Number: IPCOM000248665D
Publication Date: 2016-Dec-22
Document File: 1 page(s) / 81K

Publishing Venue

The Prior Art Database

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 100% of the total text.

Background Optimization of Speech Recognition

The problem with traditional approaches to training voice recognition systems is that asking user to repeat phrases may result in them using an un-natural (eg. over enunciated) speaking style which does not reflect everyday speech. Also people don't like taking to time to train the system - it’s an inconvenience.

Solution: The device can always be capturing speech in the background, under known 'natural' circumstances. For example, if the device is a phone, then the voice recognition system collects speech examples from conversation and thus is already optimized before the user even thinks about it.

Example: The user speaks (say, in a phone call) and a digital microphone is detecting the speech and comparing it to the input language setting of the device. If the spoken words match the language, then the speech model parameters are slowly adjusted to optimize for whatever accent and natural speaking style is present.