What May be Unique in Voice Queries?
Might a twang or a drawl influence the search results you see at Google? If you’re prone to calling an elevator a lift and tend to speak the Queen’s English in an accent similar to hers, you might see different search results than if you grew up in the Bronx or in New Orleans. If you sport a Polish accent or a Spanish one, and you perform voice searches on your phone, would receiving results in Polish or in Spanish because of your accent be a problem or a benefit? If your accent is Australian, and you search for “football” while in the US, would it surprise you to see some Australian Rules Football results returned to you?
Search engines have been using something called an Automated Search Recognition (“ASR”) engine to try to eliminate or reduce accents in voice queries by treating those as if they were noise. But the value of that noise might also be recognized as another signal that might improve search results.
A new patent was granted to Google yesterday that explores accents in voice queries in more depth. For instance, it provides this example of how a search engine might use such accent information:
By including the accent of a voice query as an input to the search result generation process, search result rankings can be biased to offer preferred results to accented voice queries.
For example, if English-language speakers with a French-language accent show a preference for a particular result for the query terms “restaurants in Scranton” than do English-language speakers with a Russian-language accent, a search engine will be more likely to select that particular result for another user who provides a voice query in the English language with a French-language accent. The preference for that particular search result would not be reflected in search results that are generated for other users that do not have that particular accent.
Such a process would be a data-driven one that could start collecting accent information when speech-based queries are made to possibly influence future search results and rankings as well. The patent tells us that accents can be indications of common backgrounds, native languages, and/or regions of origin.
The patent is:
Accent-influenced search results
Invented by Barry Hayes
US Patent 8,417,530
Granted April 9, 2013
Filed August 20, 2010
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating search results. In one aspect, a method includes obtaining a transcription of a voice query, and data that identifies an accent of the voice query, submitting the transcription and the data that identifies the accent of the voice query to a search engine to generate one or more accent-influenced results of the voice query, and providing the accent-influenced results to a client device for display.
I’ve had friends over the past 10 years or so experiment with voice recognition software with varying degrees of success. We know that Google provided a free voice search service named GOOG-411 for a number of years to collect as much data as they could about voice searches. What I hadn’t considered was that in addition to collecting data about how to filter out accents, it was also likely that they were also collecting information about those accents as well.
It seems that the noise Google was filtering out of voice queries now has the chance to be the signal that Google might use to filter search results with.