...

lib/classes/tts_engines/bark.py

@@ -194,25 +194,12 @@ class Bark(TTSUtils, TTSRegistry, name='bark'):
                             **tts_dyn_params,
                             **fine_tuned_params
                         )
-                    #audio_sentence = result.get('wav')
-                    #if is_audio_data_valid(audio_sentence):
-                    #    audio_sentence = audio_sentence.tolist()
                     if is_audio_data_valid(audio_sentence):
-                        if isinstance(audio_sentence, torch.Tensor):
-                            audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
-                        elif isinstance(audio_sentence, np.ndarray):
-                            audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        elif isinstance(audio_sentence, (list, tuple)):
-                            audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        else:
-                            error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
-                            print(error)
-                            return False
-                        if sentence[-1].isalnum() or sentence[-1] == '—':
-                            audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
+                        sourceTensor = self._tensor_type(audio_sentence)
+                        audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
                         if audio_tensor is not None and audio_tensor.numel() > 0:
+                            if sentence[-1].isalnum() or sentence[-1] == '—':
+                                audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
                             self.audio_segments.append(audio_tensor)
                             if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
                                 silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

lib/classes/tts_engines/fairseq.py

@@ -166,21 +166,11 @@ class Fairseq(TTSUtils, TTSRegistry, name='fairseq'):
                                 **speaker_argument
                             )
                     if is_audio_data_valid(audio_sentence):
-                        if isinstance(audio_sentence, torch.Tensor):
-                            audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
-                        elif isinstance(audio_sentence, np.ndarray):
-                            audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        elif isinstance(audio_sentence, (list, tuple)):
-                            audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        else:
-                            error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
-                            print(error)
-                            return False
-                        if sentence[-1].isalnum() or sentence[-1] == '—':
-                            audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
+                        sourceTensor = self._tensor_type(audio_sentence)
+                        audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
                         if audio_tensor is not None and audio_tensor.numel() > 0:
+                            if sentence[-1].isalnum() or sentence[-1] == '—':
+                                audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
                             self.audio_segments.append(audio_tensor)
                             if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
                                 silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

lib/classes/tts_engines/tacotron.py

@@ -194,21 +194,11 @@ class Tacotron2(TTSUtils, TTSRegistry, name='tacotron'):
                                 **speaker_argument
                             )
                     if is_audio_data_valid(audio_sentence):
-                        if isinstance(audio_sentence, torch.Tensor):
-                            audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
-                        elif isinstance(audio_sentence, np.ndarray):
-                            audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        elif isinstance(audio_sentence, (list, tuple)):
-                            audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        else:
-                            error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
-                            print(error)
-                            return False
-                        if sentence[-1].isalnum() or sentence[-1] == '—':
-                            audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
+                        sourceTensor = self._tensor_type(audio_sentence)
+                        audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
                         if audio_tensor is not None and audio_tensor.numel() > 0:
+                            if sentence[-1].isalnum() or sentence[-1] == '—':
+                                audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
                             self.audio_segments.append(audio_tensor)
                             if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
                                 silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

lib/classes/tts_engines/vits.py

@@ -179,21 +179,11 @@ class Vits(TTSUtils, TTSRegistry, name='vits'):
                                 **speaker_argument
                             )
                     if is_audio_data_valid(audio_sentence):
-                        if isinstance(audio_sentence, torch.Tensor):
-                            audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
-                        elif isinstance(audio_sentence, np.ndarray):
-                            audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        elif isinstance(audio_sentence, (list, tuple)):
-                            audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        else:
-                            error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
-                            print(error)
-                            return False
-                        if sentence[-1].isalnum() or sentence[-1] == '—':
-                            audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
+                        sourceTensor = self._tensor_type(audio_sentence)
+                        audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
                         if audio_tensor is not None and audio_tensor.numel() > 0:
+                            if sentence[-1].isalnum() or sentence[-1] == '—':
+                                audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
                             self.audio_segments.append(audio_tensor)
                             if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
                                 silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

lib/classes/tts_engines/xtts.py

@@ -141,12 +141,11 @@ class XTTSv2(TTSUtils, TTSRegistry, name='xtts'):
                         )
                     audio_sentence = result.get('wav')
                     if is_audio_data_valid(audio_sentence):
-                        audio_sentence = result.get('wav')
                         sourceTensor = self._tensor_type(audio_sentence)
                         audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
-                        if sentence[-1].isalnum() or sentence[-1] == '—':
-                            audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
                         if audio_tensor is not None and audio_tensor.numel() > 0:
+                            if sentence[-1].isalnum() or sentence[-1] == '—':
+                                audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
                             self.audio_segments.append(audio_tensor)
                             if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
                                 silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100

lib/classes/tts_engines/yourtts.py

@@ -109,21 +109,11 @@ class YourTTS(TTSUtils, TTSRegistry, name='yourtts'):
                             **speaker_argument
                         )
                     if is_audio_data_valid(audio_sentence):
-                        if isinstance(audio_sentence, torch.Tensor):
-                            audio_tensor = audio_sentence.detach().cpu().unsqueeze(0)
-                        elif isinstance(audio_sentence, np.ndarray):
-                            audio_tensor = torch.from_numpy(audio_sentence).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        elif isinstance(audio_sentence, (list, tuple)):
-                            audio_tensor = torch.tensor(audio_sentence, dtype=torch.float32).unsqueeze(0)
-                            audio_tensor = audio_tensor.cpu()
-                        else:
-                            error = f"{self.session['tts_engine']}: Unsupported wav type: {type(audio_sentence)}"
-                            print(error)
-                            return False
-                        if sentence[-1].isalnum() or sentence[-1] == '—':
-                            audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
+                        sourceTensor = self._tensor_type(audio_sentence)
+                        audio_tensor = sourceTensor.clone().detach().unsqueeze(0).cpu()
                         if audio_tensor is not None and audio_tensor.numel() > 0:
+                            if sentence[-1].isalnum() or sentence[-1] == '—':
+                                audio_tensor = trim_audio(audio_tensor.squeeze(), self.params['samplerate'], 0.001, trim_audio_buffer).unsqueeze(0)
                             self.audio_segments.append(audio_tensor)
                             if not re.search(r'\w$', sentence, flags=re.UNICODE) and sentence[-1] != '—':
                                 silence_time = int(np.random.uniform(0.3, 0.6) * 100) / 100