Fixed Text2Speech synthesis as NVIDIA models' URL required an update (#762)

* updated nvidia waveglow & tacotron2 model urls

* modified waveglow_handler references to nvidia models

* included waveglow requirements before creating MAR

* created requirements.txt & updated create_mar.sh

Co-authored-by: Harsh Bafna <[email protected]>
Co-authored-by: Geeta Chauhan <[email protected]>

Author: 3 people

examples/text_to_speech_synthesizer/create_mar.sh

@@ -12,7 +12,7 @@ rm -rf PyTorch
 mkdir -p PyTorch/SpeechSynthesis
 cp -r /tmp/DeepLearningExamples-torchhub/PyTorch/SpeechSynthesis/* PyTorch/SpeechSynthesis/
 zip -r tacotron.zip PyTorch
-wget https://api.ngc.nvidia.com/v2/models/nvidia/tacotron2pyt_fp32/versions/1/files/nvidia_tacotron2pyt_fp32_20190306.pth
-wget https://api.ngc.nvidia.com/v2/models/nvidia/waveglowpyt_fp32/versions/1/files/nvidia_waveglowpyt_fp32_20190306.pth
-torch-model-archiver --model-name waveglow_synthesizer --version 1.0 --model-file waveglow_model.py --serialized-file nvidia_waveglowpyt_fp32_20190306.pth --handler waveglow_handler.py --extra-files tacotron.zip,nvidia_tacotron2pyt_fp32_20190306.pth
-rm -rf nvidia_*
+wget https://api.ngc.nvidia.com/v2/models/nvidia/tacotron2_pyt_ckpt_fp32/versions/19.09.0/files/nvidia_tacotron2pyt_fp32_20190427 -O nvidia_tacotron2pyt_fp32_20190427.pth
+wget https://api.ngc.nvidia.com/v2/models/nvidia/waveglow_ckpt_fp32/versions/19.09.0/files/nvidia_waveglowpyt_fp32_20190427 -O nvidia_waveglowpyt_fp32_20190427.pth
+torch-model-archiver --model-name waveglow_synthesizer --version 1.0 --model-file waveglow_model.py --serialized-file nvidia_waveglowpyt_fp32_20190427.pth --handler waveglow_handler.py --extra-files tacotron.zip,nvidia_tacotron2pyt_fp32_20190427.pth -r requirements.txt
+rm -rf nvidia_*

examples/text_to_speech_synthesizer/requirements.txt

@@ -0,0 +1,6 @@
+numpy
+scipy
+librosa
+unidecode
+inflect
+librosa

examples/text_to_speech_synthesizer/waveglow_handler.py

@@ -38,7 +38,7 @@ def _unwrap_distributed(self, state_dict):
     def _load_tacotron2_model(self, model_dir):
         from PyTorch.SpeechSynthesis.Tacotron2.tacotron2 import model as tacotron2
         from PyTorch.SpeechSynthesis.Tacotron2.tacotron2.text import text_to_sequence
-        tacotron2_checkpoint = torch.load(os.path.join(model_dir, 'nvidia_tacotron2pyt_fp32_20190306.pth'))
+        tacotron2_checkpoint = torch.load(os.path.join(model_dir, 'nvidia_tacotron2pyt_fp32_20190427.pth'))
         tacotron2_state_dict = self._unwrap_distributed(tacotron2_checkpoint['state_dict'])
         tacotron2_config = tacotron2_checkpoint['config']
         self.tacotron2_model = tacotron2.Tacotron2(**tacotron2_config)
@@ -58,7 +58,7 @@ def initialize(self, ctx):
         with zipfile.ZipFile(model_dir + '/tacotron.zip', 'r') as zip_ref:
             zip_ref.extractall(model_dir)
 
-        waveglow_checkpoint = torch.load(os.path.join(model_dir, "nvidia_waveglowpyt_fp32_20190306.pth"))
+        waveglow_checkpoint = torch.load(os.path.join(model_dir, "nvidia_waveglowpyt_fp32_20190427.pth"))
         waveglow_state_dict = self._unwrap_distributed(waveglow_checkpoint['state_dict'])
         waveglow_config = waveglow_checkpoint['config']
         self.waveglow_model = WaveGlow(**waveglow_config)