Update app.py

app.py

@@ -1,10 +1,23 @@
 from transformers import pipeline
 import gradio as gr
+from transformers import WhisperProcessor, WhisperForConditionalGeneration
+import librosa
+import numpy as np
 
-pipe = pipeline(model="Neurai/Persian_ASR")  # change to "your-username/the-name-you-picked"
+device = 'cpu'
+processor = WhisperProcessor.from_pretrained("Neurai/Persian_ASR")
+model = WhisperForConditionalGeneration.from_pretrained("Neurai/Persian_ASR").to(device)
+forced_decoder_ids = processor.get_decoder_prompt_ids(language="fa", task="transcribe")
 
 def transcribe(audio):
-    text = pipe(audio)["text"]
+    array, sample_rate = librosa.load(audio, sr=16000,mono=True)
+    array = librosa.to_mono(array)
+    array = librosa.resample(array, orig_sr=sample_rate, target_sr=16000)
+    array = list(array)
+    input_features = processor(arrs[0:int(14*16000)], sampling_rate=16000, return_tensors="pt").input_features
+    predicted_ids = model.generate(input_features.to(device))
+    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
+    text= transcription[0]
     return text
 
 iface = gr.Interface(