Can G&L with GPU batch process?

[begeekmyfriend]

For instance, if I pass a batch of spectrograms into this function with shape of [32, ?, 1025] (in this case 32 is the batch size, 1025 is the number of FFT // 2, and they are with different frame sizes), can this funcion restructure the complete waveforms for each spectrograms?

I found the shape of the return value of tf.contrib.signal.inverse_stft is like [..., samples] but each waveforms have different lengths. therefore there are truncations for some waveforms returned from this function.