fix utf-8 error

transcribe/pipelines/base.py

@@ -59,4 +59,5 @@ class BasePipe(Process):
             if item is None:  # Check for termination signal
                 break
             out_item = self.process(item)
-            self.output_queue.put(out_item)
+            if out_item:
+                self.output_queue.put(out_item)

transcribe/pipelines/pipe_whisper.py

@@ -22,7 +22,7 @@ class WhisperPipe(BasePipe):
         source_language = in_data.source_language
         segments = self.whisper.transcribe(audio_data, source_language)
         texts = "".join([s.text for s in segments])
-        in_data.segments = [Segment(t0=s.t0, t1=s.t1, text=s.text) for s in segments]
+        in_data.segments = [Segment(t0=s.t0, t1=s.t1, text=s.text) for s in segments if s.text != "�"]
         in_data.transcribe_content = texts
         in_data.audio = b""
         return in_data

transcribe/transcription.py

@@ -176,8 +176,8 @@ class TranscriptionServer:
         frame_data = websocket.recv()
         if frame_data == b"END_OF_AUDIO":
             return False
-        # return np.frombuffer(frame_data, dtype=np.int16).astype(np.float32) / 32768.0
-        return np.frombuffer(frame_data, dtype=np.float32)
+        return np.frombuffer(frame_data, dtype=np.int16).astype(np.float32) / 32768.0
+        # return np.frombuffer(frame_data, dtype=np.float32)
 
 
     def handle_new_connection(self, websocket):

transcribe/whisper.py

@@ -2,6 +2,9 @@ from pywhispercpp.model import Model
 import soundfile
 import config
 import numpy as np
+from logging import getLogger
+
+logger = getLogger(__name__)
 
 
 class WhisperCPP:
@@ -14,6 +17,7 @@ class WhisperCPP:
             print_realtime=False,
             print_progress=False,
             print_timestamps=False,
+            translate=False
         )
         if warmup:
             self.warmup()
@@ -35,12 +39,15 @@ class WhisperCPP:
     def transcribe(self, audio_buffer:bytes, language):
         max_len, prompt = self.config_language(language)
         audio_buffer = np.frombuffer(audio_buffer, dtype=np.float32)
-        output = self.model.transcribe(
-            audio_buffer,
-            initial_prompt=prompt,
-            language=language,  
-            token_timestamps=True, 
-            max_len=max_len
-        )
-        return output
-    
+        try:
+            output = self.model.transcribe(
+                audio_buffer,
+                initial_prompt=prompt,
+                language=language,  
+                token_timestamps=True, 
+                max_len=max_len
+            )
+            return output
+        except Exception as e:
+            logger.error(e)
+            return None