Dictation accuracy boosted across languages and accents

New speech-to-text model for dictation across all plans with improved transcription accuracy across languages and accents, including multilingual speakers. Evaluations showed gains in Japanese, Korean, Chinese, Urdu, Vietnamese, accented English, and long-form Spanish, French, Italian, and Portuguese. More robust in noisy environments and with quiet/whispered speech. Word error rate at least 10% lower for top languages tested.