A PyPI port of the NVIDIA Tacotron2 model
Project description
tacotron2-model
A PyPI port of the NVIDIA Tacotron2 model
Source: https://github.com/NVIDIA/tacotron2 (model.py)
A pytorch install is required but is not added to requirements to avoid configuration issues.
The only change from the NVIDIA original is a replacement of hparams with individual arguments to remove the dependency on tf.contrib.training.HParams (deprecated since tensorflow 1).
Usage
from tacotron2_model.tacotron2 import Tacotron2
model = Tacotron2(N_MEL_CHANNELS, N_SYMBOLS, SYMBOLS_EMBEDDING_DIM,
ENCODER_N_CONVOLUTIONS, ENCODER_EMBEDDING_DIM,
ENCODER_KERNEL_SIZE, ATTENTION_RNN_DIM, ATTENTION_DIM,
ATTENTION_LOCATION_N_FILTERS, ATTENTION_LOCATION_KERNEL_SIZE,
DECODER_RNN_DIM, PRENET_DIM, MAX_DECODER_STEPS, GATE_THRESHOLD,
P_ATTENTION_DROPOUT, P_DECODER_DROPOUT, POSTNET_EMBEDDING_DIM,
POSTNET_KERNEL_SIZE, POSTNET_N_CONVOLUTIONS).cuda()
print(model.eval())
Example params
N_MEL_CHANNELS = 80
N_SYMBOLS = 148
SYMBOLS_EMBEDDING_DIM = 512
ENCODER_N_CONVOLUTIONS = 3
ENCODER_EMBEDDING_DIM = 512
ENCODER_KERNEL_SIZE = 5
ATTENTION_RNN_DIM = 1024
ATTENTION_DIM = 128
ATTENTION_LOCATION_N_FILTERS = 32
ATTENTION_LOCATION_KERNEL_SIZE = 31
DECODER_RNN_DIM = 1024
PRENET_DIM = 256
MAX_DECODER_STEPS = 1000
GATE_THRESHOLD = 0.5
P_ATTENTION_DROPOUT = 0.1
P_DECODER_DROPOUT = 0.1
POSTNET_EMBEDDING_DIM = 512
POSTNET_KERNEL_SIZE = 5
POSTNET_N_CONVOLUTIONS = 5
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
tacotron2-model-0.1.4.tar.gz
(7.6 kB
view hashes)
Built Distribution
Close
Hashes for tacotron2_model-0.1.4-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 36b99dc6e3c20f034e90ffb48005fe02a38e245e4ea43f457531566b67cfbb3a |
|
MD5 | 4ed1c42f092d6865b5ac3ef2b45397cc |
|
BLAKE2b-256 | 39d00b31fbe60624894f589a46877d1631ab109cb9016b6a112e38381112cd74 |