fix small bug

app.py

@@ -8,6 +8,23 @@ from PIL import Image, ImageDraw, ImageFont
 from transformers import pipeline
 
 
+# checkpoint = "openai/whisper-tiny"
+# checkpoint = "openai/whisper-base"
+checkpoint = "openai/whisper-small"
+
+# We need to set alignment_heads on the model's generation_config (at least
+# until the models have been updated on the hub).
+# If you're going to use a different version of whisper, see the following
+# for which values to use for alignment_heads:
+# https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a
+
+# whisper-tiny
+# alignment_heads = [[2, 2], [3, 0], [3, 2], [3, 3], [3, 4], [3, 5]]
+# whisper-base
+# alignment_heads = [[3, 1], [4, 2], [4, 3], [4, 7], [5, 1], [5, 2], [5, 4], [5, 6]]
+# whisper-small
+alignment_heads = [[5, 3], [5, 9], [8, 0], [8, 4], [8, 7], [8, 8], [9, 0], [9, 7], [9, 9], [10, 5]]
+
 max_duration = 60  # seconds
 fps = 25
 video_width = 640
@@ -22,10 +39,6 @@ font = ImageFont.truetype("Lato-Regular.ttf", 40)
 text_color = (255, 200, 200)
 highlight_color = (255, 255, 255)
 
-# checkpoint = "openai/whisper-tiny"
-# checkpoint = "openai/whisper-base"
-checkpoint = "openai/whisper-small"
-
 if torch.cuda.is_available() and torch.cuda.device_count() > 0:
     from transformers import (
         AutomaticSpeechRecognitionPipeline,
@@ -45,18 +58,12 @@ if torch.cuda.is_available() and torch.cuda.device_count() > 0:
 else:
     pipe = pipeline(model=checkpoint)
 
-# TODO: no longer need to set these manually once the models have been updated on the Hub
-# whisper-tiny
-# pipe.model.generation_config.alignment_heads = [[2, 2], [3, 0], [3, 2], [3, 3], [3, 4], [3, 5]]
-# whisper-base
-# pipe.model.generation_config.alignment_heads = [[3, 1], [4, 2], [4, 3], [4, 7], [5, 1], [5, 2], [5, 4], [5, 6]]
-# whisper-small
-pipe.model.generation_config.alignment_heads = [[5, 3], [5, 9], [8, 0], [8, 4], [8, 7], [8, 8], [9, 0], [9, 7], [9, 9], [10, 5]]
+pipe.model.generation_config.alignment_heads = alignment_heads
 
 chunks = []
 
 start_chunk = 0
-last_draws = []
+last_draws = None
 last_image = None
 
 
@@ -126,7 +133,7 @@ def predict(audio_path):
     global chunks, start_chunk, last_draws, last_image
 
     start_chunk = 0
-    last_draws = []
+    last_draws = None
     last_image = None
 
     audio_data, sr = librosa.load(audio_path, mono=True)