azraelkuan/tensorflow_wavenet_vocoder

wavenet vocoder using tensorflow

Hi there 👋

tensorflow_wavenet_vocoder's Issues

there are some question about my implementation

Hi azraelkuan, thanks for your work and sharing!
I encounter three questions during implementation.

my environment: win10, python==3.6.7, tensorflow==1.11, anaconda==3

one is when I finished "preprocess.py", my file(LJSpeech-1-mel.npy) was only 176KB and file(LJSpeech-1-audio.npy) was only 281KB. I think it may be that something is repeatedly covered or the problem is caused by the difference between windows and Linux? i am not sure about this.

The second problem is encountered during the Synthesize step. I didn't find a file called "eavl.txt". There are only three files (LJSpeech-1-audio.npy, LJSpeech-1-mel.npy, train.txt) in output path after preprocess.

the final question is it seems that the input of the parameter is adjusted, unlike the command in the readme?
about '--eval_txt' i just set the output folder for preprocess.

tensorflow_wavenet_vocoder>python mul_generate.py --eval_txt ./FeaPath/ --wav_out_path ./WavOut/ checkpoint ./log_ljspeech/train/2018-11-18T18-07-48/model.ckpt-99999 ---hparams gc_enable=False,global_channel=0,global_cardinality=0,NPY_DATAROOT=/your_npy_datadir/,sample_rate=22050 usage: mul_generate.py [-h] [--logdir LOGDIR] [--temperature TEMPERATURE] [--save_every SAVE_EVERY] [--eval_txt EVAL_TXT] [--hparams HPARAMS] checkpoint mul_generate.py: error: unrecognized arguments: --wav_out_path checkpoint ./log_ljspeech/train/2018-11-18T18-07-48/model.ckpt-99999 ---hparams gc_enable=False,global_channel=0,global_cardinality=0,NPY_DATAROOT=/your_npy_datadir/,sample_rate=22050

or this code can't running on windows?
Tell me if I'm wrong, thanks ^_^

missing the last segment of local condition

s = np.random.randint(0, len(local_condition) - max_frames)

--> s = np.random.randint(0, len(local_condition) - max_frames +1)

How many steps of the iteration can have acceptable results

Thanks for sharing the great work.
I have trained on Chinese data for 48000 steps, but now I can only get noise through the mul-gengerate.py.And I found that the loss first decrease and then increase.Can you give me some advise.Thanks.

how to increase the batch size?

Hi azraelkuan!
when i using your wavenet vocoder in my corpus, keeping other parameter unchanged. but the loss does not converge.
my computer is 2080ti with 11 GB's memory, but my batch size can only be 1. when i directly increase it to 2, its memory overflows.
how could i increase my batch size ?
thanks!

环境问题

请问不用gpu 是否也可以训练

there are some question about my implementation，please help me

Exception in thread Thread-1:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError
Exception in thread Thread-2:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError

Exception in thread Thread-8:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError

Exception in thread Thread-7:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError

Exception in thread Thread-6:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError

Exception in thread Thread-4:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError

Exception in thread Thread-5:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError

Exception in thread Thread-3:
Traceback (most recent call last):
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 916, in _bootstrap_inner
self.run()
File "D:\Anaconda3\envs\wtx\lib\threading.py", line 864, in run
self._target(*self._args, **self._kwargs)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 145, in thread_main
assert_ready_for_upsampling(wav, local_condition)
File "D:\tf-wavenet_vocoder-master\apps\vocoder\datasets\data_feeder.py", line 50, in assert_ready_for_upsampling
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size()
AssertionError

Could you please share some trained params?

Dear Kuan Chen,
Could you please share your checkpoints for the LJspeech and CMU_arctic, as well as the corresponding hparams settings? That will be much helpful for me.
Thank you in advance!

Songxiang Liu

Do your project generate raw speech samples from acoustic features, or from text, like TTS?

I found this project is trying to generate raw speech samples from acoustic features like previous vocoders (WORLD,DIO,etc.)
https://github.com/r9y9/wavenet_vocoder

But it seems your project is to generate waveform and conditioned by text files?

It's input might different with the traditional one, right?

Thanks

ignore this