Update app.py

app.py

@@ -15,7 +15,9 @@ processor = AutoProcessor.from_pretrained(model_id)
 model = Wav2Vec2ForCTC.from_pretrained(model_id)
 model = BetterTransformer.transform(model)
 
-def transcribe(audio_file_mic=None, audio_file_upload=None, language="English (eng)"):
+pipe = pipeline("automatic-speech-recognition", model=model)
+
+def transcribe(audio_file_mic=None, audio_file_upload=None):
     if audio_file_mic:
         audio_file = audio_file_mic
     elif audio_file_upload:
@@ -28,11 +30,6 @@ def transcribe(audio_file_mic=None, audio_file_upload=None, language="English (e
     if sample_rate != 16000:
         speech = librosa.resample(speech, orig_sr=sample_rate, target_sr=16000)
 
-    # Keep the same model in memory and simply switch out the language adapters by calling load_adapter() for the model and set_target_lang() for the tokenizer
-    language_code = iso_codes[language]
-    processor.tokenizer.set_target_lang(language_code)
-    pipe = pipeline("automatic-speech-recognition", model=model)
-
     return pipe(audio_file)["text"]