basve

3 ms·

1. 1.▲
   by basve 10y ago

   Afaik none of the released libraries support the TTS experiment described in the paper. Deepmind used pre-computed linguistic features to guide the system in generating natural sounding speech, so your output will probably depend on the qua
2. 2.▲
   by basve 10y ago

   My intuition is that the doubling up to 512 does increase the receptive field, but you're essentially building a non-linear convolutional filter with a kernel size of 1024. The network benefits from stacking multiple of these groups, b
3. 3.▲
   by basve 10y ago

   90 minutes for 1s of audio was reported by someone from Google on twitter, but the tweet has been deleted. I've clarified in the readme that my measurements are for a much lighter/smaller model than Deepmind's :).
4. 4.▲
   by basve 10y ago

   And I should add that this was measured using a downsized model (just two blocks of dilated convolutions and a sampling rate of 4khz). Deepmind's paper does not report how many stacks are used to generate the samples, but I assume it&#
5. 5.▲
   WaveNet implementation in Keras(github.com)

   106 points by basve 10y ago|15 comments