load faster

app.py

@@ -1,3 +1,5 @@
+import os
+
 import numpy as np
 import unicodedata
 import diff_match_patch as dmp_module
@@ -40,7 +42,9 @@ def style_text(diff):
     return fullText
 
 
-dataset = load_dataset("distil-whisper/tedlium-long-form", split="validation")
+dataset = load_dataset(
+    "distil-whisper/tedlium-long-form", split="validation", num_proc=os.cpu_count()
+)
 
 csv = pd.read_csv("assets/large-v2.csv")
 
@@ -76,7 +80,9 @@ def get_visualisation(idx):
 
 
 if __name__ == "__main__":
-    gr.Markdown("Analyse the transcriptions generated by the Whisper large-v2 model on the TEDLIUM dev set.")
+    gr.Markdown(
+        "Analyse the transcriptions generated by the Whisper large-v2 model on the TEDLIUM dev set."
+    )
     with gr.Blocks() as demo:
         slider = gr.Slider(
             minimum=1, maximum=len(norm_target), step=1, label="Dataset sample"
@@ -85,8 +91,12 @@ if __name__ == "__main__":
         audio_out = gr.Audio(label="Audio input")
         with gr.Row():
             wer = gr.Number(label="WER")
-            relative_insertions = gr.Number(label="Relative insertions (# insertions / target length)")
-            relative_length = gr.Number(label="Relative length (reference length / target length)")
+            relative_insertions = gr.Number(
+                label="Relative insertions (# insertions / target length)"
+            )
+            relative_length = gr.Number(
+                label="Relative length (reference length / target length)"
+            )
         text_out = gr.Markdown(label="Text difference")
 
         btn.click(