initial commit

transcription/transcription.py

@@ -0,0 +1,57 @@
+import whisper
+
+
+# clojure for model
+def model_loader() -> callable:
+    """
+    Clojure for loading the model
+    :return: function that loads the model
+    """
+
+    model = None
+
+    def load_model(config: dict):
+        """
+        Load the model
+        :return: model
+        """
+        nonlocal model
+
+        if model is not None:
+            return model
+        else:
+            model_type = config['model']['type']
+            model = whisper.load_model(model_type)
+            return model
+
+    return load_model
+
+
+load_model = model_loader()
+
+
+def transcribe_audio(audio_file_path: str,
+                     output_file_path: str,
+                     config: dict) -> str:
+    """
+    Transcribe an audio file with openai whisper
+    :param audio_file_path: filepath of the audio file
+    :param output_file_path: filepath of the output file
+    :param config: app config
+    :return: transcribed text
+    """
+
+    # load the model
+    model = load_model(config)
+
+    # move model to desired device
+    desired_device = config['model']['device']
+    model.to(desired_device)
+
+    result = model.transcribe(audio_file_path)
+    result_text = result['text']
+
+    # write the result to a text file
+    with open(output_file_path, 'w') as result_file:
+        result_file.write(result["text"])
+    return result_text