Update src/streamlit_app.py

src/streamlit_app.py

@@ -220,23 +220,49 @@ with tab5:
 
 with tab6: 
     st.header("Results: WER vs Dataset Size")
-    
+
     st.write("""
-        Overall, the WER decreases as the number of training hours increases across all models and languages. This trend underscores the importance of dataset size in improving ASR performance. However, the rate of improvement varies significantly between models, with some benefiting more from additional data than others.    
-        """)
+    Overall, the Word Error Rate (WER) decreases as the number of training hours increases across all models and languages.
+    This highlights the importance of dataset size in improving ASR performance, although the rate of improvement varies
+    significantly between models.
+    """)
 
     # XLS-R
     st.subheader("XLS-R")
-    st.image("Images/xlsrlog.png", caption="Log WER vs Training Hours for XLS-R")
+    st.write("""
+    XLS-R shows a steep decline in log WER as the dataset size increases, especially in low-to-moderate data regimes.
+    The improvement slows as the dataset becomes larger, suggesting diminishing returns in high-data settings.
+    """)
+    st.image("src/Images/xlsrlog.png", caption="Log WER vs Training Hours for XLS-R")
 
     # W2v-BERT
     st.subheader("W2v-BERT")
-    st.image("Images/bertlog.png", caption="Log WER vs Training Hours for W2v-BERT")
+    st.write("""
+    W2v-BERT exhibits a more gradual decline in log WER. It performs well in low-data settings, showing stable reduction
+    in WER as dataset size increases. This makes it suitable for low-resource languages.
+    """)
+    st.image("src/Images/bertlog.png", caption="Log WER vs Training Hours for W2v-BERT")
 
     # Whisper
     st.subheader("Whisper")
-    st.image("Images/whisperlog.png", caption="Log WER vs Training Hours for Whisper")
+    st.write("""
+    Whisper shows a consistent but moderate decline in log WER. Improvements are more linear compared to XLS-R, benefiting
+    steadily from additional data, but it does not reach XLS-R’s high-data performance.
+    """)
+    st.image("src/Images/whisperlog.png", caption="Log WER vs Training Hours for Whisper")
 
     # MMS
     st.subheader("MMS")
-    st.image("Images/mmslog.png", caption="Log WER vs Training Hours for MMS")
+    st.write("""
+    MMS shows significant improvement between 1–5 hours of training across multiple languages. However, the rate of
+    improvement declines as more data is added. MMS performs strongly in both low- and high-data settings.
+    """)
+    st.image("src/Images/mmslog.png", caption="Log WER vs Training Hours for MMS")
+
+    # Overall Insight
+    st.subheader("Overall Insights")
+    st.write("""
+    - All models exhibit the largest WER improvements when training data is scarce.
+    - Beyond a certain dataset size, adding more data results in marginal gains.
+    - Dataset size remains a critical factor, but its impact plateaus once the model is trained on sufficient data.
+    """)