It seems that it is easy for companies to implement different languages to their virtual assistants but it is not so easy if we look at Google Assistant that only speaks English or Alexa among others. Apple received a lot of criticism at the time of its launch for not being available in all languages and the attendees of the competition are also being criticized for this - although it seems that they criticize them a little less - it seems that this is not so simple. language learning process and Siri is already fluent in 36 languages.
How does Siri learn different languages?
In the latest version released by Apple, the company announced the arrival of another language, the Shanghainese. Well, these languages are being learned little by little and the process is quite curious. The first and most remarkable thing is that in the case of Apple it requires people who speak directly the language they want to add to the assistant, these people will read various paragraphs, sentences, words with different accents and all of this is recorded. All these raw recordings will be recorded again by other people from the same country but from other locations to define accents and how to vocalize.
All these recordings are entered into a algorithmic machine training model that predicts and selects words that are not in its database, in addition it is putting words together and adding all the content, shaping phrases and others. Once this process is done Apple includes language in iOS and macOS dictation, so that they continue to learn in addition to storing those same words with real noise, coughing, pauses of the different users when speaking, etc ... Then it will be used to improve these phrases and voice database.
And with all that database in the form of "new" and much larger voices that Apple already has in its possession, it only remains to pass everything learned to Siri and do it clean. For that, what they do in Cupertino is to filter the voices and repeat the recording process with users for Siri to synthesize audio by correcting imperfections from text to speech when asked by a user.
A curious and not at all simple process explained by Alex Steel at Reuters, from which we expect improvements once artificial intelligence is really used and not previously recorded questions and answers. Evidently Siri works great, but there is always room for improvement and SiriKit will also add functions in third-party applications, which undoubtedly make Siri one of the best voice assistants.
