从新提交到gitee 仓库

rag/llm/tts_model.py

@@ -70,6 +70,8 @@ class Base(ABC):
     def tts(self, audio):
         pass
 
+    def end_tts(self):
+        pass
     def normalize_text(self, text):
         return re.sub(r'(\*\*|##\d+\$\$|#)', '', text)
 
@@ -118,15 +120,30 @@ class FishAudioTTS(Base):
 class QwenTTS(Base):
     def __init__(self, key, model_name, base_url=""):
         import dashscope
-
+        print("---begin--init QwenTTS--")  # cyx
         self.model_name = model_name
         dashscope.api_key = key
-
+        self.synthesizer = None
+        self.callback = None
+        self.is_cosyvoice = False
+        self.voice = ""
+        if '/' in model_name:
+            parts = model_name.split('/', 1)
+            # 返回分离后的两个字符串parts[0], parts[1]
+            if parts[0] == 'cosyvoice-v1':
+                self.is_cosyvoice = True
+                self.voice = parts[1]
     def tts(self, text):
         from dashscope.api_entities.dashscope_response import SpeechSynthesisResponse
-        from dashscope.audio.tts import ResultCallback, SpeechSynthesizer, SpeechSynthesisResult
-        from collections import deque
-
+        if self.is_cosyvoice is False:
+            from dashscope.audio.tts import ResultCallback, SpeechSynthesizer, SpeechSynthesisResult
+            from collections import deque
+        else:
+            # cyx 2025 01 19 测试cosyvoice  使用tts_v2 版本
+            from dashscope.audio.tts_v2 import ResultCallback, SpeechSynthesizer, AudioFormat  #, SpeechSynthesisResult
+            from dashscope.audio.tts import  SpeechSynthesisResult
+            from collections import deque
+        # print(f"--QwenTTS--tts_stream begin-- {text}")  # cyx
         class Callback(ResultCallback):
             def __init__(self) -> None:
                 self.dque = deque()
@@ -149,6 +166,7 @@ class QwenTTS(Base):
                 self.dque.append(None)
 
             def on_error(self, response: SpeechSynthesisResponse):
+                print("Qwen tts error", str(response))
                 raise RuntimeError(str(response))
 
             def on_close(self):
@@ -158,20 +176,91 @@ class QwenTTS(Base):
                 if result.get_audio_frame() is not None:
                     self.dque.append(result.get_audio_frame())
 
+
+
+            # --------------------------
+
+        class Callback_v2(ResultCallback):
+            def __init__(self) -> None:
+                self.dque = deque()
+
+            def _run(self):
+                while True:
+                    if not self.dque:
+                        time.sleep(0)
+                        continue
+                    val = self.dque.popleft()
+                    if val:
+                        yield val
+                    else:
+                        break
+
+            def on_open(self):
+                pass
+
+            def on_complete(self):
+                self.dque.append(None)
+
+            def on_error(self, response: SpeechSynthesisResponse):
+                print("Qwen tts error", str(response))
+                raise RuntimeError(str(response))
+
+            def on_close(self):
+                print("---Qwen call back close")  # cyx
+                pass
+            """ canceled for test 语音大模型CosyVoice
+            def on_event(self, result: SpeechSynthesisResult):
+                if result.get_audio_frame() is not None:
+                    self.dque.append(result.get_audio_frame())
+            """
+
+            def on_event(self, message):
+                # print(f"recv speech synthsis message {message}")
+                pass
+            # 以下适合语音大模型CosyVoice
+            def on_data(self, data: bytes) -> None:
+                if len(data) > 0:
+                    self.dque.append(data)
+
+
+            # --------------------------
         text = self.normalize_text(text)
-        callback = Callback()
-        SpeechSynthesizer.call(model=self.model_name,
-                               text=text,
-                               callback=callback,
-                               format="mp3")
+
+
         try:
-            for data in callback._run():
+            #if self.model_name != 'cosyvoice-v1':
+            if self.is_cosyvoice is False:
+                self.callback = Callback()
+                SpeechSynthesizer.call(model=self.model_name,
+                                       text=text,
+                                       callback=self.callback,
+                                       format="mp3")
+            else:
+                self.callback = Callback_v2()
+                self.synthesizer = SpeechSynthesizer(
+                    model='cosyvoice-v1',
+                    # voice="longyuan", #"longfei",
+                    voice = self.voice,
+                    callback=self.callback,
+                    format=AudioFormat.MP3_44100HZ_MONO_256KBPS
+                )
+
+                self.synthesizer.call(text)
+        except Exception as e:
+            print(f"---dale---20  error {e}")  # cyx
+        # -----------------------------------
+        try:
+            for data in self.callback._run():
                 yield data
+            print(f"---Qwen return data {num_tokens_from_string(text)}")
             yield num_tokens_from_string(text)
 
         except Exception as e:
-            raise RuntimeError(f"**ERROR**: {e}")
+           raise RuntimeError(f"**ERROR**: {e}")
 
+    def end_tts(self):
+        if self.synthesizer:
+            self.synthesizer.streaming_complete()
 
 class OpenAITTS(Base):
     def __init__(self, key, model_name="tts-1", base_url="https://api.openai.com/v1"):