Fix windows model loading by restructuring the code

Makefile

@@ -5,7 +5,7 @@ run:
 	pip3 install --force-reinstall .
 
 run-e2e-tests:
-	pip install --force-reinstall ctranslate2==3.13.0
+	pip install --force-reinstall ctranslate2==3.14.0
 	pip install --force-reinstall faster-whisper==0.6.0
 	CT2_USE_MKL="False" CT2_FORCE_CPU_ISA='GENERIC' nose2 -s e2e-tests
 

src/whisper_ctranslate2/live.py

@@ -57,6 +57,7 @@ def __init__(
         self.speaking = False
         self.blocks_speaking = 0
         self.buffers_to_process = []
+        self.transcribe = None
 
     @staticmethod
     def is_available():
@@ -119,17 +120,21 @@ def process(self):
             if self.verbose:
                 print("\n\033[90mTranscribing..\033[0m")
 
-            result = Transcribe().inference(
+            if not self.transcribe:
+                self.transcribe = Transcribe(
+                    self.model_path,
+                    self.device,
+                    self.device_index,
+                    self.compute_type,
+                    self.threads,
+                    self.cache_directory,
+                    self.local_files_only,
+                )
+
+            result = self.transcribe.inference(
                 audio=_buffer.flatten(),
-                model_path=self.model_path,
-                cache_directory=self.cache_directory,
-                local_files_only=self.local_files_only,
                 task=self.task,
                 language=self.language,
-                threads=self.threads,
-                device=self.device,
-                device_index=self.device_index,
-                compute_type=self.compute_type,
                 verbose=self.verbose,
                 live=True,
                 options=self.options,

src/whisper_ctranslate2/transcribe.py

@@ -94,23 +94,17 @@ def _get_vad_parameters_dictionary(self, options):
 
         return vad_parameters
 
-    def inference(
+    def __init__(
         self,
-        audio: Union[str, BinaryIO, np.ndarray],
         model_path: str,
-        cache_directory: str,
-        local_files_only: bool,
-        task: str,
-        language: str,
-        threads: int,
         device: str,
         device_index: Union[int, List[int]],
         compute_type: str,
-        verbose: bool,
-        live: bool,
-        options: TranscriptionOptions,
+        threads: int,
+        cache_directory: str,
+        local_files_only: bool,
     ):
-        model = WhisperModel(
+        self.model = WhisperModel(
             model_path,
             device=device,
             device_index=device_index,
@@ -120,9 +114,18 @@ def inference(
             local_files_only=local_files_only,
         )
 
+    def inference(
+        self,
+        audio: Union[str, BinaryIO, np.ndarray],
+        task: str,
+        language: str,
+        verbose: bool,
+        live: bool,
+        options: TranscriptionOptions,
+    ):
         vad_parameters = self._get_vad_parameters_dictionary(options)
 
-        segments, info = model.transcribe(
+        segments, info = self.model.transcribe(
             audio=audio,
             language=language,
             task=task,

src/whisper_ctranslate2/whisper_ctranslate2.py

@@ -513,18 +513,21 @@ def main():
 
         return
 
+    transcribe = Transcribe(
+        model_dir,
+        device,
+        device_index,
+        compute_type,
+        threads,
+        cache_directory,
+        local_files_only,
+    )
+
     for audio_path in audio:
-        result = Transcribe().inference(
+        result = transcribe.inference(
             audio_path,
-            model_dir,
-            cache_directory,
-            local_files_only,
             task,
             language,
-            threads,
-            device,
-            device_index,
-            compute_type,
             verbose,
             False,
             options,