Hi,
After successful training of duration and acoustic models, I tried to synthesize my voice. After running 07_run_merlin.sh, I get the following error when trying to synthesize:
TypeError: Cannot convert Type TensorType(float32, 3D) (of Variable HostFromGpu(gpuarray).0) into Type TensorType(float32, matrix). You can try to manually convert HostFromGpu(gpuarray).0 into a TensorType(float32, matrix).
Acoustic model has architecture [512, 512], ['LSTM', 'LSTM'], but I would not say that is the problem because another model with architecture [256, 256], ['LSTM', 'LSTM'] (so the only difference is the number of hidden units) succesfuly synthesized my txt files.
Does anyone know what the problem could be? I already added floatX=float32 to submit.sh script, and it still doesn't work.
Thanks!
Hi, After successful training of duration and acoustic models, I tried to synthesize my voice. After running 07_run_merlin.sh, I get the following error when trying to synthesize:
TypeError: Cannot convert Type TensorType(float32, 3D) (of Variable HostFromGpu(gpuarray).0) into Type TensorType(float32, matrix). You can try to manually convert HostFromGpu(gpuarray).0 into a TensorType(float32, matrix).
Acoustic model has architecture [512, 512], ['LSTM', 'LSTM'], but I would not say that is the problem because another model with architecture [256, 256], ['LSTM', 'LSTM'] (so the only difference is the number of hidden units) succesfuly synthesized my txt files.
Does anyone know what the problem could be? I already added floatX=float32 to submit.sh script, and it still doesn't work. Thanks!