single file

app.py

@@ -2,15 +2,34 @@
 import gradio as gr
 from whisper import generate
 from AinaTheme import theme
+from faster_whisper import WhisperModel
+import torch
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+torch_dtype = "float32"
 
 MODEL_NAME = "Systran/faster-whisper-large-v3"
+print("Loading model ...")
+model = WhisperModel(MODEL_NAME, compute_type=torch_dtype)
+print("Loading model done.")
 
 def transcribe(inputs):
     if inputs is None:
         raise gr.Error("Cap fitxer d'àudio introduit! Si us plau pengeu un fitxer "\
                        "o enregistreu un àudio abans d'enviar la vostra sol·licitud")
 
-    return generate(audio_path=inputs)
+    segments, _ = model.transcribe(
+        inputs, 
+        # language="ca",
+        # chunk_length=30,
+        task="transcribe",
+        word_timestamps=False,
+    )
+
+    text = ""
+    for segment in segments:
+        text += " " + segment.text.strip()
+    return text
 
 
 description_string = "Transcripció automàtica de micròfon o de fitxers d'àudio.\n Aquest demostrador s'ha desenvolupat per"\

whisper.py

@@ -1,24 +0,0 @@
-from faster_whisper import WhisperModel
-import torch
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-torch_dtype = "float32"
-
-MODEL_NAME = "Systran/faster-whisper-large-v3"
-model = WhisperModel(MODEL_NAME, compute_type=torch_dtype)
-
-def generate(audio_path):
-    #check audio lenght
-    segments, _ = model.transcribe(
-        audio_path, 
-        # language="ca",
-        # chunk_length=30,
-        task="transcribe",
-        word_timestamps=False,
-    )
-
-    text = ""
-    for segment in segments:
-        text += " " + segment.text.strip()
-    return text
-