Last week I wrote comprehensively on the topic of speech recognition, detailing both how it pertains to interactive voice response and its greater technological capabilities overall. In fact, I lauded Siri as the probable future for all voice response technology, and noted the vast amount of funding and research went into its development.
The technology that powers Siri is truly groundbreaking, and is currently unmatched by any other speech recognition and processing tools available. This is not to say, however, that Siri is foolproof.
Yesterday as I was driving along, I heard a piece on the program All Things Considered on NPR. The station did a story on Siri’s ability to understand people speaking a Siri-recognized language (which as it stands is exclusively English, German, or French).
The results were quite amusing, and demonstrated some of the potential challenges that the global market presents to the software. NPR asked Neil McIntosh, a native of Scotland who has lived in the US for the better part of 14 years, to test out Siri’s functionality when presented with accented English.
Siri was able to identify about half of the questions it was being asked correctly, but some of the resulting misunderstandings were very entertaining. Here is an excerpt from a transcript of the conversation (please keep in mind that McIntosh is speaking in a Scottish accent. You can hear it for yourself here:
MCINTOSH: Remind me to call the vet.
COMPUTER VOICE: I don’t know what you mean by, I need powders.
The Scottish accent wasn’t the only speech that Siri had a difficult time identifying. It was hard for Siri to understand individuals who spoke in Scottish English, but also in Boston and Japanese accents as well.
It is worth noting that Siri comes equipped with an option for specifying geographical location, with the user having the ability to select from various options. When switched toe the United Kingdom setting however, Siri was even less effective than under American English.
Per Apple’s website: “ Since every language has its own accents and dialects, the accuracy rate will be higher for native speakers.” So does Siri not consider English spoken in a Scottish accent English then?
It is of course a bit more complicated than that. For Siri to be able to understand and acclimate itself to various local, national, and international accents, it would have to be able to be operate on a regional basis.
Per NPR, regional accents can vary by even 30 or 40 miles, so the likelihood of Siri being adapted and tailored on a geographical basis is improbable if not impossible. With Apple, however, I think many of us have been pleasantly surprised to learn that the sky is the limit, and this could be the latest and greatest step they take in their speech recognition research and development.
