add package requirements

InferenceInterfaces/Meta_FastSpeech2.py

@@ -1,9 +1,7 @@
-import itertools
 import os
 
 import librosa.display as lbd
 import matplotlib.pyplot as plt
-import sounddevice
 import soundfile
 import torch
 
@@ -75,46 +73,3 @@ class Meta_FastSpeech2(torch.nn.Module):
             plt.subplots_adjust(left=0.05, bottom=0.1, right=0.95, top=.9, wspace=0.0, hspace=0.0)
             plt.show()
         return wave
-
-    def read_to_file(self, text_list, file_location, silent=False, dur_list=None, pitch_list=None, energy_list=None):
-        """
-        :param silent: Whether to be verbose about the process
-        :param text_list: A list of strings to be read
-        :param file_location: The path and name of the file it should be saved to
-        """
-        if not dur_list:
-            dur_list = []
-        if not pitch_list:
-            pitch_list = []
-        if not energy_list:
-            energy_list = []
-        wav = None
-        silence = torch.zeros([24000])
-        for (text, durations, pitch, energy) in itertools.zip_longest(text_list, dur_list, pitch_list, energy_list):
-            if text.strip() != "":
-                if not silent:
-                    print("Now synthesizing: {}".format(text))
-                if wav is None:
-                    if durations is not None:
-                        durations = durations.to(self.device)
-                    if pitch is not None:
-                        pitch = pitch.to(self.device)
-                    if energy is not None:
-                        energy = energy.to(self.device)
-                    wav = self(text, durations=durations, pitch=pitch, energy=energy).cpu()
-                    wav = torch.cat((wav, silence), 0)
-                else:
-                    wav = torch.cat((wav, self(text, durations=durations.to(self.device), pitch=pitch.to(self.device), energy=energy.to(self.device)).cpu()), 0)
-                    wav = torch.cat((wav, silence), 0)
-        soundfile.write(file=file_location, data=wav.cpu().numpy(), samplerate=48000)
-
-    def read_aloud(self, text, view=False, blocking=False):
-        if text.strip() == "":
-            return
-        wav = self(text, view).cpu()
-        wav = torch.cat((wav, torch.zeros([24000])), 0)
-        if not blocking:
-            sounddevice.play(wav.numpy(), samplerate=48000)
-        else:
-            sounddevice.play(torch.cat((wav, torch.zeros([12000])), 0).numpy(), samplerate=48000)
-            sounddevice.wait()

packages.txt.py

@@ -0,0 +1,2 @@
+espeak-ng
+libsndfile