Model size is now a choice.

app.py

@@ -6,7 +6,6 @@ from simple_diarizer.diarizer import Diarizer
 import streamlit as st
 import base64
 
-model_size = 'small'
 
 
 
@@ -16,6 +15,7 @@ def create_download_link(val, filename, label):
 
 
 def speech_to_text(uploaded):
+    st.write(f'Using {model_size} model.')
     model = whisper.load_model(model_size)
     result = model.transcribe(uploaded,verbose=True)
     return f'You said: {result["text"]}'
@@ -59,7 +59,7 @@ def transcribe(uploaded, nu_speakers):
     audio_bytes = audio_file.read()
     st.audio('mono.wav', format='audio/wav')
 
-    with st.spinner(text="Transcribing..."):
+    with st.spinner(text=f"Transcribing using {model_size} model..."):
         tdf = audio_to_df(uploaded)
     with st.spinner(text="Segmenting..."):
         sdf = segment(nu_speakers)
@@ -107,10 +107,21 @@ st.markdown(descript)
 form = st.form(key='my_form')
 uploaded = form.file_uploader("Choose a file")
 nu_speakers = form.slider('Number of speakers in recording:', min_value=1, max_value=8, value=2, step=1)
+models = form.selectbox(
+    'Which Whisper model?',
+    ('Tiny (fast)', 'Base (good)', 'Small (great but slow)'), index=1)
+
 submit = form.form_submit_button("Transcribe!")
 
 
 if submit:
+    if models == 'Tiny (fast)':
+        model_size = 'tiny'
+    elif models == 'Base (good)':
+        model_size ='base'
+    elif models == 'Small (great but slow)':
+        model_size = 'small'
+
     bytes_data = uploaded.getvalue()
     with open('temp_audio', 'wb') as outfile:
         outfile.write(bytes_data)