Nowadays, speech spoofing is so common that it presents a great challenge to social security. Thus, it is of great significance to recognize a spoofed speech from a genuine one. Most of the current researches have focused on voice conversion (VC), synthesis and recapture which mimic a target speaker to break through ASV systems by increased false acceptance rates. However, there exists another type of spoofing, voice transformation (VT), that transforms a speech signal without a target in order ‘not to be recognized’ by increased false reject rates. VT has received much less attention. Thus, in this paper, we investigate the model of VT and propose a method using a very deep dense convolutional network with 135 layers to detect VT spoofed speeches from genuine speeches. The experimental results show that the average accuracies over intra-database and cross-database outperform the reported state-of-the-art methods.