Update app.py

app.py

@@ -394,38 +394,27 @@ def offline_run(files, batch_size: int, want_ts: bool):
 with gr.Blocks(title="Parakeet-TDT v3: Streaming (Mic) + Offline (File)") as demo:
     gr.Markdown(
     """
-    # FINALLY, SIMPLE EXPLANATION OF THE NVIDIA NEMO TECHNICALS! (tl;dr scroll down for kinda working realtime transcript and file upload and transcript)
-    app itself is "not great not terrible" but if you have "the WILL" you will copy paste,
-    the code into AI of your choice and develop from there, good luck! (same goes for me :P)
+    # FINALLY, SIMPLE EXPLANATION OF THE NVIDIA NEMO TECHNICALS!
     
-    This is a CHAD (smol/basic) version of the idea of local transcription in real-time
-    on cheap hardware, for example to use with rapsberry pi locally to "do this, do that"
-    (you can easily use another SMOL LLM to in real-time take an action when you say "open the door",
-    or any other stuff that you can imagine, its a modern programming - whre the steps are defined by language
-    and the context - no more silly programming language syntax :D)
+    This is a CHAD (smol one, not a GIGA one/basic) version of the idea of local transcription in real-time
+    on cheap hardware, for example to use with rapsberry pi locally to say - do this, do that
+    - you can easily use another SMOL LLM to in real-time take an action when you say "open the door
     
-    # Beam ASR = Google Maps for Speech - THE ANALOGY TO THE NVIDIA TECHNICAL TERMS USED IN THIS APP
-    NOTE: this app is not using lame chunking of audio like others... We are fully compatible here with modern
-    architecture of Parakeet-TDT-v3 model, its no joke anymore - soon, the "cache- streaming" will be
-    implemented - another gamechanger, anyway - THE WORK IS ONGOING ON THIS APP, FOR TODAY HERE YOU HAVE SMOL CHAD COMMIT XD
+    # Beam ASR = Google Maps for Speech - THE ANALOGY OF USED FEATURES TO THE NVIDIA TECHNICAL TERMS USED IN THIS APP
+    NOTE this app is not using lame chunking of audio like others... We are fully compatible here with modern
+    architecture of Parakeet-TDT-v3 model, its no joke anymore - soon, the cache-aware streaming will be
+    implemented - another gamechanger, anyway - THE WORK IS ONGOING ON THIS APP, FOR TODAY HERE YOU HAVE SMOL CHAD COMMIT
     
     **One-route vs many**  
     - Greedy: pick the first route and drive.  
     - Beam: keep several good routes, update with traffic, follow the best.
+    
     **Beam size**  
     - How many alternate routes you watch at once.
+    
     **Label-looping**  
     - Make back-to-back turns at the same intersection when signs are clear (e.g., "right, then immediate merge"). Faster, fewer stutters.
-    **Fair scoring (length-norm)**  
-    - Compare routes by ETA, not just distance. A longer route does not win just for having more segments.
-    **Blank / stability**  
-    - The "wait" option at an intersection. If signals are unclear, pause instead of taking a wrong turn.
-    **Streaming (mic)**  
-    - You only see a few blocks ahead (chunk + right-context). The planner keeps alternates and snaps overlapping segments so the path does not repeat or skip.
-    **File (batch)**  
-    - Whole trip is visible. Same multi-route logic in one pass. No stitching needed.
-    **Trade-off**  
-    - More routes watched -> better choices, higher planning cost. Beam 8 is a solid balance.
+    
     """
     )