search

barronalex / Tacotron

Implementation of Google's Tacotron in TensorFlow
236 stars 80 forks source link

promising results #4

Open zhang-jian opened 7 years ago

zhang-jian commented 7 years ago

Thank you, very promising result. Dataset: arctic only Parameters: everything is default Training time: 14 hours

untitled

individualAudio.zip

YisenWang commented 7 years ago

@barronalex @zhang-jian Have you run into this issues when using arctic?

Traceback (most recent call last): File "train.py", line 117, in train(model, config) File "train.py", line 28, in train batch_inputs = data_input.build_dataset(sess, inputs) File "/data1/home/yisenwang/TTS/tacotron2/Tacotron/data_input.py", line 26, in build_dataset dataset = tf.contrib.data.Dataset.from_tensor_slices(placeholders) File "/data1/home/yisenwang/TTS/tacotron2/Tacotron/lib/python2.7/site-packages/tensorflow/contrib/data/python/ops/dataset_ops.py", line 473, in from_tensor_slices return TensorSliceDataset(tensors) File "/data1/home/yisenwang/TTS/tacotron2/Tacotron/lib/python2.7/site-packages/tensorflow/contrib/data/python/ops/dataset_ops.py", line 907, in init for i, t in enumerate(nest.flatten(tensors)) File "/data1/home/yisenwang/TTS/tacotron2/Tacotron/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 610, in convert_to_tensor as_ref=False) File "/data1/home/yisenwang/TTS/tacotron2/Tacotron/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 675, in internal_convert_to_tensor ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref) File "/data1/home/yisenwang/TTS/tacotron2/Tacotron/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 921, in _autopacking_conversion_function return _autopacking_helper(v, inferred_dtype, name or "packed") File "/data1/home/yisenwang/TTS/tacotron2/Tacotron/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 863, in _autopacking_helper "%s to %s (Tensor is: %r)" % (elem.dtype, dtype, elem)) TypeError: Cannot convert a list containing a tensor of dtype <dtype: 'float32'> to <dtype: 'int32'> (Tensor is: <tf.Tensor 'Placeholder_2:0' shape=(1131, 64, 5125) dtype=float32>)

auzxb commented 7 years ago

Check the version of your Tensorflow. Whether it's tensorflow_gpu-1.2.0 or tensorflow_cpu-1.2.0 ?@YisenWang

YisenWang commented 7 years ago

It is already 1.2.0. @auzxb

YisenWang commented 7 years ago

I have fixed this one when I update tensorflow from 1.2.0rc2 to 1.2.0

jacksmithe commented 7 years ago

@zhang-jian For some reason I can't hear your audio samples. Maybe the files are buggie?

zhang-jian commented 7 years ago

@jacksmithe https://www.dropbox.com/s/845j6yll7uxf0bt/individualAudio.wav?dl=0

onyedikilo commented 7 years ago

@zhang-jian is this sample generated through test.py or train.py?

mrelich commented 7 years ago

Anyone else having difficulty getting consistent output during training and from test.py? I am finding test.py gives garbage while output during training is getting better after many epochs.