feat: 后端封装tts，可以通过配置文件切换openvoice和vits

utils/tts/openvoice_utils.py

@@ -164,11 +164,14 @@ class TextToSpeech:
         """
         return audio_data.cpu().detach().float().numpy()
     
-    def numpy2bytes(self, audio_data: np.ndarray):
-        """
-        numpy类型转bytes
-        """
-        return (audio_data*32768.0).astype(np.int32).tobytes()
+    def numpy2bytes(self, audio_data):
+        if isinstance(audio_data, np.ndarray):
+            if audio_data.dtype == np.dtype('float32'):
+                audio_data = np.int16(audio_data * np.iinfo(np.int16).max)
+            audio_data = audio_data.tobytes()
+            return audio_data
+        else:
+            raise TypeError("audio_data must be a numpy array")
 
     def _base_tts(self,
                  text: str,
@@ -292,14 +295,11 @@ class TextToSpeech:
     
     def synthesize(self,
             text: str,
-            sdp_ratio=0.2,
-            noise_scale=0.6,
-            noise_scale_w=0.8,
-            speed=1.0,
-            quite=True,
-            
+            tts_info,
             source_se: Optional[np.ndarray]=None,
             target_se: Optional[np.ndarray]=None,
+            sdp_ratio=0.2,
+            quite=True,
             tau :float=0.3,
             message :str="default"):
         """
@@ -316,11 +316,11 @@ class TextToSpeech:
         """
         audio, sr = self._base_tts(text, 
                                    sdp_ratio=sdp_ratio,
-                                   noise_scale=noise_scale,
-                                   noise_scale_w=noise_scale_w,
-                                   speed=speed,
+                                   noise_scale=tts_info['noise_scale'],
+                                   noise_scale_w=tts_info['noise_scale_w'],
+                                   speed=tts_info['speed'],
                                    quite=quite)
-        if self.use_tone_convert:
+        if self.use_tone_convert and target_se.size>0:
             tts_sr = self.base_tts_model.hps.data.sampling_rate
             converter_sr = self.tone_color_converter.hps.data.sampling_rate
             audio = F.resample(audio, tts_sr, converter_sr)
@@ -342,4 +342,5 @@ class TextToSpeech:
             save_path: 保存路径
         """
         sf.write(save_path, audio, sample_rate)
-        print(f"Audio saved to {save_path}")
+        print(f"Audio saved to {save_path}")
+        

utils/tts/vits_utils.py

@@ -2,6 +2,7 @@ import os
 import numpy as np
 import torch
 from torch import LongTensor
+from typing import Optional
 import soundfile as sf
 # vits
 from .vits import utils, commons
@@ -79,19 +80,19 @@ class TextToSpeech:
             print(f"Synthesis time: {time.time() - start_time} s")
         return audio
     
-    def synthesize(self, text, language, speaker_id, noise_scale, noise_scale_w, length_scale, save_audio=False, return_bytes=False):
+    def synthesize(self, text, tts_info,target_se: Optional[np.ndarray]=None, save_audio=False, return_bytes=True):
         if not len(text):
             return "输入文本不能为空！", None
         text = text.replace('\n', ' ').replace('\r', '').replace(" ", "")
         if len(text) > 100 and self.limitation:
             return f"输入文字过长！{len(text)}>100", None
-        text = self._preprocess_text(text, language)
-        audio = self._generate_audio(text, speaker_id, noise_scale, noise_scale_w, length_scale)
+        text = self._preprocess_text(text, tts_info['language'])
+        audio = self._generate_audio(text, tts_info['speaker_id'], tts_info['noise_scale'], tts_info['noise_scale_w'], tts_info['length_scale'])
         if self.debug or save_audio:
             self.save_audio(audio, self.RATE, 'output_file.wav')
         if return_bytes:
             audio = self.convert_numpy_to_bytes(audio)
-        return self.RATE, audio
+        return audio, self.RATE
 
     def convert_numpy_to_bytes(self, audio_data):
         if isinstance(audio_data, np.ndarray):