Open erogol opened 6 years ago
@erogol
I test the inference on a P40 GPU with the default configuration(i.e., n_layers=16, mu_quantization=256, sampling_rate=16000, implementation=PERSISTENT
).
It takes about 3 to 5 seconds to synthesis a 9 seconds long audio.
So I think nv-wavenet is real-time at least for these parameters.
@nanma thx for the values.
Are there any measurements for the inference time? Should I expect real-time or close to real-time? It'd be great too see some values.