add model size option from server

whisper_live/server.py

@@ -101,9 +101,10 @@ class TranscriptionServer:
             multilingual=options["multilingual"],
             language=options["language"],
             task=options["task"],
-            client_uid=options["uid"]
+            client_uid=options["uid"],
+            model_size=options["model_size"]
         )
-
+        
         self.clients[websocket] = client
         self.clients_start_time[websocket] = time.time()
 
@@ -127,7 +128,8 @@ class TranscriptionServer:
 
             except Exception as e:
                 logging.error(e)
-                self.clients[websocket].cleanup()
+                if self.clients[websocket].model_size is not None:
+                    self.clients[websocket].cleanup()
                 self.clients.pop(websocket)
                 self.clients_start_time.pop(websocket)
                 logging.info("Connection Closed.")
@@ -180,7 +182,16 @@ class ServeClient:
     SERVER_READY = "SERVER_READY"
     DISCONNECT = "DISCONNECT"
 
-    def __init__(self, websocket, task="transcribe", device=None, multilingual=False, language=None, client_uid=None):
+    def __init__(
+        self,
+        websocket,
+        task="transcribe",
+        device=None,
+        multilingual=False,
+        language=None,
+        client_uid=None,
+        model_size="small"
+        ):
         """
         Initialize a ServeClient instance.
         The Whisper model is initialized based on the client's language and device availability.
@@ -199,11 +210,23 @@ class ServeClient:
         self.client_uid = client_uid
         self.data = b""
         self.frames = b""
-        self.language = language if multilingual else "en"
+        self.model_sizes = [
+            "tiny", "base", "small", "medium", "large-v2"
+        ]
+        self.multilingual = multilingual
+        self.model_size = self.get_model_size(model_size)
+        self.language = language if self.multilingual else "en"
         self.task = task
+        self.websocket = websocket
+        
         device = "cuda" if torch.cuda.is_available() else "cpu"
+        
+        
+        if self.model_size == None:
+            return
+        
         self.transcriber = WhisperModel(
-            "small" if multilingual else "small.en", 
+            self.model_size, 
             device=device,
             compute_type="int8" if device=="cpu" else "float16", 
             local_files_only=False,
@@ -228,7 +251,6 @@ class ServeClient:
         self.pick_previous_segments = 2
 
         # threading
-        self.websocket = websocket
         self.trans_thread = threading.Thread(target=self.speech_to_text)
         self.trans_thread.start()
         self.websocket.send(
@@ -240,34 +262,30 @@ class ServeClient:
             )
         )
     
-    def fill_output(self, output):
+    def get_model_size(self, model_size):
         """
-        Format the current incomplete transcription output by combining it with previous complete segments.
+        Returns the whisper model size based on multilingual.
-        The resulting transcription is wrapped into two lines, each containing a maximum of 50 characters.
+        """
-
+        if model_size not in self.model_sizes:
-        It ensures that the combined transcription fits within two lines, with a maximum of 50 characters per line.
+            self.websocket.send(
-        Segments are concatenated in the order they exist in the list of previous segments, with the most
+                json.dumps(
-        recent complete segment first and older segments prepended as needed to maintain the character limit.
+                    {
-        If a 3-second pause is detected in the previous segments, any text preceding it is discarded to ensure
+                        "uid": self.client_uid,
-        the transcription starts with the most recent complete content. The resulting transcription is returned
+                        "status": "ERROR",
-        as a single string.
+                        "message": f"Invalid model size {model_size}. Available choices: {self.model_sizes}"
-
+                    }
-        Args:
+                )
-            output(str): The current incomplete transcription segment.
+            )
+            return None
         
-        Returns:
+        if model_size == "large-v2":
-            str: A formatted transcription wrapped in two lines.
+            self.multilingual = True
-        """
+            return model_size
-        text = ''
+
-        pick_prev = min(len(self.text), self.pick_previous_segments)
+        if not self.multilingual:
-        for seg in self.text[-pick_prev:]:
+            model_size = model_size + ".en"
-            # discard everything before a 3 second pause
+
-            if seg == '':
+        return model_size
-                text = ''
-            else:
-                text += seg
-        wrapped = "".join(text + output)
-        return wrapped
     
     def add_frames(self, frame_np):
         """