missTwo blind graders agree (D7, rule D16). Not in the audit sample (D11).
Grader A: miss
Synthetic voices in 2009 (concatenative TTS) were intelligible but clearly machine-like. Even Google's WaveNet in 2016, a large step forward, was still less convincing than human speech. Fully human-sounding voices came only after 2014.
Test: Synthetic speech indistinguishable from human speech: not met in 2009 or by 2014.
WaveNet (Wikipedia) (2016): As of 2016, WaveNet synthesis outperformed earlier TTS but was still less convincing than actual human speech.
Grader B: miss
2009 synthetic voices matched humans only in narrow, tuned contexts and were recognizably synthetic in general use. Broadly natural speech arrived with neural TTS (WaveNet 2016, Tacotron 2 2018).
Test: Capability test: fully human-sounding general TTS not achieved by 2014.
Wikipedia: Speech synthesis (accessed 2026-09-28): Best unit-selection systems 'often indistinguishable' from humans only in contexts they are tuned for; WaveNet 2016 and Tacotron 2 2018 brought broadly natural speech.
All captured fields
rank
42
kind
decade_prediction
section
Education
subject
Synthetic voices sound fully human
target year
2009
timing
in
position
Chapter 9, "2009"; item Education 16 in Kurzweil 2010, p. 41
confidence
high
note
Wording as reproduced by the author in "How My Predictions Are Faring" (2010); book page not given there. Subject label is the essay's own item heading.