While many AI researchers suggest that Siri is lagging behind competitors such as Google Assistant and Amazon Alexa, one area in which Apple’s voice assistant clearly has the lead over its competitors is in its extremely wide language support. A new report from Reuters notes that Siri can speak and understand 21 different languages, with localized dialect support for over 36 countries in total. By comparison, Microsoft Cortana supports only eight languages for 13 countries, Google’s Assistant speaks four, and Amazon’s Alexa is limited to only English and German.
Reuters also notes that Siri will also be adding support for Shanghainese, a special dialect of Wu Chinese spoken only around Shanghai.
Alex Acero, the head of Apple’s speech team for Siri, explained to Reuters the process Apple uses in adding new language support to Siri. Acero notes that they begin by bringing in people to read passages in a range of accents and dialects, and then transcribe those passages by hand to provide an exact representation that the system can compare to and learn from. A range of sounds and variety of voices is used to improve recognition quality, and then a language model is built to try and predict typical word sequences.
Apple then deploys “dictation mode” to capture a small percentage of anonymized audio recordings, complete with background noise and mumbled words, which are transcribed by humans, resulting in about a 50 percent reduction in the rate of errors in speech recognition. One sufficient data has been gathered, Apple records a voice actor to speak for Siri in the new language, and then releases the new language with a list of answers to what Apple expects to be the most common questions. Siri then learns during real-world use about what users ask, and Apple tweaks Siri on an ongoing basis to refine the process and improve the quality of speech recognition and answers, pushing out updates about every two weeks.
