WaveNet: A Generative Model for Raw Audio

12 Sep 2016Aaron van den OordSander DielemanHeiga ZenKaren SimonyanOriol VinyalsAlex GravesNal KalchbrennerAndrew SeniorKoray Kavukcuoglu

This paper introduces WaveNet, a deep neural network for generating raw audio waveforms. The model is fully probabilistic and autoregressive, with the predictive distribution for each audio sample conditioned on all previous ones; nonetheless we show that it can be efficiently trained on data with tens of thousands of samples per second of audio... (read more)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Speech Synthesis Mandarin Chinese WaveNet (L+F) Mean Opinion Score 4.08 # 1
Speech Synthesis Mandarin Chinese HMM-driven concatenative Mean Opinion Score 3.47 # 3
Speech Synthesis Mandarin Chinese LSTM-RNN parametric Mean Opinion Score 3.79 # 2
Speech Synthesis North American English WaveNet (L+F) Mean Opinion Score 4.21 # 3
Speech Synthesis North American English HMM-driven concatenative Mean Opinion Score 3.86 # 5
Speech Synthesis North American English LSTM-RNN parametric Mean Opinion Score 3.67 # 6