Application desing

src/streamlit_app.py

@@ -131,7 +131,7 @@ def handle_upload(label: str, key: str) -> np.ndarray | None:
         return None
 
 def delta(x):
-    """Computes first-order difference along time axis."""
+    """Computes first-order derivative along time axis."""
     return x[:, 1:] - x[:, :-1]
 
 def array_to_spectrogram(audio_np: np.ndarray,
@@ -175,9 +175,9 @@ def verify_speakers(model, audio_left, audio_right, margin):
     cosine_similarity = float(cosine_similarity.numpy().squeeze())
 
     if cosine_similarity >= margin:
-        st.success("Both voice recordings belong to the same person.")
+        st.success("Both utterances belong to the same speaker.")
     else:
-        st.warning("The voice recordings belong to different people.")
+        st.warning("The utterances are from different speakers.")
     st.caption(f"Cosine similarity: {cosine_similarity:.4f}, margin: {margin:.4f}")
 
 # ========= Load model =========
@@ -188,7 +188,7 @@ if st.session_state.load_model_button:
             filename="best_model.keras",
             revision="v1.0.0",
         )
-        st.success("Model loaded. You can now upload/record audio files.")
+        st.success("Model loaded — you can upload audio files or record utterances.")
         st.download_button(
             "(Option) Download the model file",
             data=model_bytes,
@@ -201,8 +201,8 @@ if st.session_state.load_model_button:
     left_column, right_column = st.columns(2)
 
     with left_column:
-        st.subheader("Left input")
-        record_left = st.checkbox("Record left input")
+        st.subheader("Voice Sample 1")
+        record_left = st.checkbox("Record first voice sample", key="chk_record_left")
         if record_left:
             audio_left = handle_record("Record (left)")
         else:
@@ -210,12 +210,12 @@ if st.session_state.load_model_button:
         if audio_left is not None:
             st.session_state.audio_left = audio_left
             fig = plot_waveform(audio_left, FS, "Left audio waveform")
-            st.pyplot(fig, use_container_width=True)
+            st.pyplot(fig, width="stretch")
             st.caption(f"Samples: {audio_left.size}  •  Duration: {audio_left.size/FS:.2f}s")
 
     with right_column:
-        st.subheader("Right input")
-        record_right = st.checkbox("Record right input")
+        st.subheader("Voice Sample 2")
+        record_right = st.checkbox("Record second voice sample", key="chk_record_right")
         if record_right:
             audio_right = handle_record("Record (right)")
         else:
@@ -223,12 +223,12 @@ if st.session_state.load_model_button:
         if audio_right is not None:
             st.session_state.audio_right = audio_right
             fig = plot_waveform(audio_right, FS, "Right audio waveform")
-            st.pyplot(fig, use_container_width=True)
+            st.pyplot(fig, width="stretch")
             st.caption(f"Samples: {audio_right.size}  •  Duration: {audio_right.size/FS:.2f}s")
 
     if audio_left is not None and audio_right is not None:
         margin = st.slider('Selected margin:', -1.0, 1.0, 0.26, 0.01)
-        verify_button = st.button("Verify speaker!")
+        verify_button = st.button("Verify Speaker")
         if verify_button:
             try:
                 verify_speakers(model, audio_left, audio_right, margin)