update examples

gradio_asr_en_libri100_word_vs_bpe.py

@@ -48,10 +48,11 @@ def predict(wav_file):
     B = model_bpe(speech)[0]
     word_decoded = W[0]
     bpe_decoded = B[0]
-    comparison = ''.join(list(d.compare([word_decoded+'\n'], [bpe_decoded+'\n'])))
+    # comparison = ''.join(list(d.compare([word_decoded+'\n'], [bpe_decoded+'\n'])))
 
     logger.info('predicted')
-    return word_decoded, bpe_decoded, comparison
+    # return word_decoded, bpe_decoded, comparison
+    return word_decoded, bpe_decoded
 
 iface = gr.Interface(
     predict,
@@ -63,7 +64,7 @@ iface = gr.Interface(
     outputs=[
         gr.outputs.Textbox(label='Decoding result (word-token model)'),
         gr.outputs.Textbox(label='Decoding result (BPE-token model)'),
-        gr.outputs.Textbox(label='Comparison'),
+        # gr.outputs.Textbox(label='Comparison'),
     ],
     examples=examples,
     # article='<p style="text-align:center">Model URL<a target="_blank" href="https://huggingface.co/jkang/espnet2_librispeech_100_conformer">🤗</a></p>',

test.py

@@ -0,0 +1,12 @@
+def best_overlap(a, b):
+    return max([(score(a[offset:], b), offset) for offset in range(len(a))], key=lambda x: x[0])[1]
+
+def score(a, b):
+    return sum([a[i] == b[i] for i in range(len(a))])
+
+a = 'abcde'
+b = 'abbde'
+
+print(best_overlap(a, b))
+print(a + '-' * best_overlap(a, b))
+print('-' * best_overlap(a, b) + b)