feature: 集成bert_vits

utils/bert_vits2_utils.py

@@ -395,29 +395,28 @@ class TextToSpeech:
     
     def synthesize(self,
                    text,
-                   speaker_idx=0,       # self.speakers 的 index，指定说话
-                   sdp_ratio=0.5,
-                   noise_scale=0.6,
-                   noise_scale_w=0.9,
-                   length_scale=1.0,    # 越大语速越慢
-                   language="mix",      # ["ZH", "EN", "mix"] 三选一
-                   opt_cut_by_send=False, # 按句切分    在按段落切分的基础上再按句子切分文本
-                   interval_between_para=1.0, # 段间停顿(秒)，需要大于句间停顿才有效
-                   interval_between_sent=0.2, # 句间停顿(秒)，勾选按句切分才生效
-                   audio_prompt=None,
-                   text_prompt="",
-                   prompt_mode="Text prompts",
-                   style_text="",        #    "使用辅助文本的语意来辅助生成对话（语言保持与主文本相同）\n\n"
-                                         # "**注意**：不要使用**指令式文本**（如：开心），要使用**带有强烈情感的文本**（如：我好快乐！！！）\n\n"
-                                         # "效果较不明确，留空即为不使用该功能"
-                   style_weight=0.7,     # "主文本和辅助文本的bert混合比率，0表示仅主文本，1表示仅辅助文本
-                   en_ratio=1.0          # 中英混合时，英文速度控制，越大英文速度越慢 
+                   tts_info, 
                    ):
         """
         return: audio, sample_rate
         """
+        speaker_id = tts_info['speaker_id'] # self.speakers 的 index，指定说话
+        sdp_ratio = tts_info['sdp_ratio']
+        noise_scale = tts_info['noise_scale']
+        noise_scale_w = tts_info['noise_scale_w']
+        length_scale = tts_info['length_scale']
+        language = tts_info['language'] # ["ZH", "EN", "mix"] 三选一
+        opt_cut_by_send = tts_info['opt_cut_by_send'] 
+        interval_between_para = tts_info['interval_between_para'] # 段间停顿(秒)，需要大于句间停顿才有效
+        interval_between_sent = tts_info['interval_between_sent'] # 句间停顿(秒)，勾选按句切分才生效
+        audio_prompt = None
+        text_prompt = ""
+        prompt_mode = "Text prompts"
+        style_text = tts_info['style_text']
+        style_weight = tts_info['style_weight']
+        en_ratio = tts_info['en_ratio']
         
-        speaker = self.speakers[speaker_idx]
+        speaker = self.speakers[speaker_id]
 
         if language == "mix":
             language, text = self.format_utils(text, speaker)
@@ -455,9 +454,17 @@ class TextToSpeech:
                 style_weight
             )
         
-        # return text_output, audio_output
-        return audio_output[1], audio_output[0]
+        return self.convert_numpy_to_bytes(audio_output[1])
     
     def print_speakers_info(self):
         for i, speaker in enumerate(self.speakers):
             print(f"id: {i}, speaker: {speaker}")
+            
+    def convert_numpy_to_bytes(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")