finish description

app.py

@@ -177,16 +177,25 @@ if __name__ == "__main__":
         )
         gr.Markdown(
             """
-            Analyse the transcriptions generated by the Whisper and Distil-Whisper models on the TED-LIUM dev set. 
-            Analysis is performed on the overall level, where statistics are computed over the entire dev set, and also a per-sample level.
-            The transcriptions for both models are shown at the bottom of the demo. The text diff for each is computed
-            relative to the target transcriptions, where insertions are displayed in <span style='background-color:Lightgreen'>green</span>, and 
-            deletions in <span style='background-color:#FFCCCB'><s>red</s></span>.
-            
+            One of the major claims of the <a href="https://arxiv.org/abs/2311.00430"> Distil-Whisper paper </a> is that 
+            that Distil-Whisper hallucinates less than Whisper on long-form audio. To demonstrate this, we'll analyse the
+            transcriptions generated by <a href="https://huggingface.co/openai/whisper-large-v2"> Whisper </a> 
+            and <a href="https://huggingface.co/distil-whisper/distil-large-v2"> Distil-Whisper </a> on the 
+            <a href="https://huggingface.co/datasets/distil-whisper/tedlium-long-form"> TED-LIUM </a> validation set.
+
             To quantify the amount of repetition and hallucination in the predicted transcriptions, we measure the number 
-            of repeated 5-gram word duplicates (5-Dup.) and the insertion error rate (IER). Overall, Distil-Whisper has 
-            roughly half the number of 5-Dup. and IER. This indicates that it has a lower propensity to hallucinate
-            compared to the Whisper model.
+            of repeated 5-gram word duplicates (5-Dup.) and the insertion error rate (IER). Analysis is performed on the 
+            overall level, where statistics are computed over the entire dataset, and also a per-sample level (i.e. an 
+            on an individual example basis).
+            
+            The transcriptions for both models are shown at the bottom of the demo. We compute a text difference for each 
+            relative to the ground truth transcriptions. Insertions are displayed in <span style='background-color:Lightgreen'>green</span>,
+            and deletions in <span style='background-color:#FFCCCB'><s>red</s></span>. Multiple words in <span style='background-color:Lightgreen'>green</span>
+            indicates that a model has hallucinated, since it has inserted words not present in the ground truth transcription.
+            
+            Overall, Distil-Whisper has roughly half the number of 5-Dup. and IER. This indicates that it has a lower 
+            propensity to hallucinate compared to the Whisper model. Try both models with some of the TED-LIUM examples 
+            and view the reduction in hallucinations for yourself!
             """
         )
         gr.Markdown("**Overall statistics:**")