Artificial intelligence · 2016
WaveNet
A network generates raw audio sample by sample, and synthetic voices sound human.
DeepMind's WaveNet used dilated causal convolutions to model audio waveforms at 16,000 samples per second, halving the gap to human speech naturalness. It shipped in Google Assistant in 2017 and started neural text-to-speech.
What it built on 1
What it led to 1
Related measurements
The dashed line marks when it first worked.
Training compute of notable AI models Source: Epoch AI, Notable AI Models (CC BY 4.0)
ML hardware FLOP/s per dollar Source: Epoch AI, ML hardware (CC BY 4.0)