search

TensorSpeech / TensorFlowTTS

:stuck_out_tongue_closed_eyes: TensorFlowTTS: Real-Time State-of-the-art Speech Synthesis for Tensorflow 2 (supported including English, French, Korean, Chinese, German and Easy to adapt for other languages)
https://tensorspeech.github.io/TensorFlowTTS/
Apache License 2.0
3.8k stars 810 forks source link

optimizer further by using fake-quantize aware and pruning #47

Closed gongchenghhu closed 4 years ago

gongchenghhu commented 4 years ago

Thanks for your this great job. I have trained Tacotron2 model as your repository. Now I am trying to convert our model to a int8 quantization model with Tensorflow Lite. But I encountered some errors，when I use converter = tf.lite.TFLiteConverter.from_saved_model("./test_saved") or converter = tf.lite.TFLiteConverter.from_keras_model(tacotron2). And do you have any advise about optimizer further by using fake-quantize aware and pruning? Thank you very much.

dathudeptrai commented 4 years ago

@gongchenghhu hi, it's not ez to convert Tacotron2 to tflite, need to write custom C++ ops. Even you can convert and run it on low-device, it's still cann't run real time :(. I suggest you use FastSpeech instead. For fake-quantize aware and pruning, you can refer official doc here (https://www.tensorflow.org/model_optimization/guide/quantization/training_comprehensive_guide). I don't really know how much tensorflow support fake-quantize and pruning for LSTM, but if model is just consist of Convolution, everything would be ok, i think. BTW, how was a quality of ur tacotron ?

gongchenghhu commented 4 years ago

@dathudeptrai Thanks for your reply. I will try the fake-quantize aware and pruning (https://www.tensorflow.org/model_optimization/guide/quantization/training_comprehensive_guide)

manmay-nakhashi commented 4 years ago

@gongchenghhu @dathudeptrai i am also trying to convert fastspeech model to Tflite and it is crashing in length regulator

manmay-nakhashi commented 4 years ago

@gongchenghhu @dathudeptrai you can follow this issue for TFLite conversion, https://github.com/tensorflow/tensorflow/issues/40504 , can you also tell help me with this ?

gongchenghhu commented 4 years ago

@manmay-nakhashi ok, I will see your issue.

dathudeptrai commented 4 years ago

@gongchenghhu @manmay-nakhashi sorry :)). Hope Tensorflower can help you in this case :D

manmay-nakhashi commented 4 years ago

@dathudeptrai haha :P

manmay-nakhashi commented 4 years ago

@gongchenghhu i was able to convert melgan to tflite , i think we can try to convert Fastspeeh2 to tflite , structurally model code seems to be easier then tacotron2 and fastspeech @dathudeptrai do you think that would be good addition to repo ?

sujeendran commented 4 years ago

@manmay-nakhashi Could it be possible for you to share how you managed to convert melgan to tflite? Is the tflite model capable of inferencing?

aidansmyth95 commented 3 months ago

Do we have an example of FastSpeech2 conversion to tflite int8?