Update app.py

app.py

@@ -1,9 +1,6 @@
 import streamlit as st
 from st_audiorec import st_audiorec
 
-import librosa
-import soundfile
-
 from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
 from datasets import load_dataset
 import torch
@@ -14,26 +11,29 @@ audio_transcription: str = None
 
 def main ():
     
+    print("Run init model")
+    pipe = init_model()
     print("Run init model")
     pipe = init_model()
     # x = st.slider('Select a value')
     # st.write(x, 'squared is', x * x)
 
+    print("Render UI")
     print("Render UI")
     wav_audio_data = st_audiorec()
 
     if wav_audio_data is not None:
+        print("Loading data...")
         st.audio(wav_audio_data, format='audio/wav')
+        sample = transcribe(wav_audio_data, pipe)
+        st.write('Sample:', sample)
+        
 
-    print("Load data: audio1")
+    
     # dataset = load_dataset("distil-whisper/librispeech_long", "clean", split="validation")
     # sample = dataset[0]["audio"]
     
-    audio_file_path = "data/audio1.wav"
-    audio_data, sample_rate = librosa.load(audio_file_path)
-    sample = transcribe(audio_data, pipe)
-    st.write('Sample:', transcribe(sample))
-
+    # audio_file_path = "data/audio1.wav"
 
 def init_model ():
     device = "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -62,7 +62,10 @@ def init_model ():
     )
     print(f'Init model successful: {model}' )
     return pipe
+    print(f'Init model successful: {model}' )
+    return pipe
 
+def transcribe (audio_sample: bytes, pipe) -> str:
 def transcribe (audio_sample: bytes, pipe) -> str:
 
     # dataset = load_dataset("distil-whisper/librispeech_long", "clean", split="validation")