ValueError: Cannot feed value of shape (256, 9, 129) for Tensor 'model/inputs:0', which has shape

问题描述

• speech-enhancement工程 [https://github.com/jtkim-kaist/Speech-enhancement]
• 使用环境
  - tensorflow 1.7
  - librosa
  - matlab2017a
  - tensorboardX

当我运行main.py时，发生以下错误。此时，使用的模型是fnn。 后来，当我在trnmodle.py中检查fnn模型时，我发现了错误的原因。 但是，我没有修改datareader.py中的参数或命令。当我使用lstm模型时发生了同样的错误。

ValueError: Cannot feed value of shape (256, 9, 129, 1) for Tensor ‘model/inputs:0’, which has shape ‘(?, 129, 1, 1)’

调用的fnn模型的为 config.py ,如下所示：

	import numpy as np
	import os

	mode = "fnn"  # fnn, fcn, lstm, sfnn, irm, tsn

	if mode == 'lstm':
		 time_width = int(16)
	else:
		time_width = int(9)

	fs = float(8000)
	win_size = int(0.025 * fs)  # The number of samples in window
	win_step = int(0.010 * fs)
	# nfft = np.int(2 ** (np.floor(np.log2(win_size) + 1)))
	nfft = np.int(256)

	freq_size = int(nfft/2+1)

	lr = 0.0001
	lrDecayRate = .99  # 0.99
	lrDecayFreq = 2000

	keep_prob = 0.9
	global_std = 1.18

	device = '/gpu:0'