Can TTS(text to speech) model be inferened on the VPU Plugin?

[Roxy1]

Hello, My model framework is PyTorch, converted to ONNX and then to an IR file, but it cannot be inferenced on the VPU. Is it not possible to run a TTS model on the VPU?

[Peh_Intel]

Hi Roxy1,

For your information, we do have Public Pre-Trained Text-to-Speech models, ForwardTacotron and WaveRNN models.

Referring to Public Pre-Trained Models Device Support, ForwardTacotron model is not supported by MYRIAD while WaveRNN model is supported by MYRIAD.

Next, I try inferencing these models with Text-to-speech Python Demo. This demo requires ForwardTacotron and WaveRNN models.

python text_to_speech_demo.py -i test.txt --model_duration forward-tacotron-duration-prediction.xml --model_forward forward-tacotron-regression.xml --model_rnn wavernn-rnn.xml --model_upsample wavernn-upsampler.xml -d MYRIAD

WaveRNN model can be loaded to MYRIAD plugin, but ForwardTacotron model is failed to be loaded due to unsupported layer by MYRIAD.

Regards,

Peh

[Peh_Intel]

Hi Roxy1,

This thread will no longer be monitored since we have provided answer. If you need any additional information from Intel, please submit a new question. 

Regards,

Peh