Open Radar

 
Area:
AVFoundation

Summary:
Often, but not always, the small versions of the Japanese characters や、ゆ、よ are mispronounced by AVSpeechSynthesizer as if they were the large versions. A few examples of this are

はっしゃ which is pronounced as はっしや
なっしゃ which is pronounced as なっしや
じんりょく which is pronounced as じんりよく

I'm sure there are more cases, but these are reliably reproducible examples.

Steps to Reproduce:
let synthesizer = AVSpeechSynthesizer()
let utterance = AVSpeechUtterance(string: "なっしゃ")
utterance.voice = AVSpeechSynthesisVoice(language: "ja-JP")
synthesizer.speak(utterance)


Expected Results:
The text should be pronounced "nassha"

Actual Results:
The text is pronounced "nasshiya"

Version:
iOS 10.2

Notes:
This is reproducible both in the simulator and on an actual device.

Configuration:
iPhone SE