Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from jiwer import wer, compute_measures | |
| # Function to highlight errors | |
| def highlight_errors(ground_truth, hypothesis): | |
| measures = compute_measures(ground_truth, hypothesis) | |
| highlighted_hyp = [] | |
| # Split the ground truth and hypothesis into words | |
| gt_words = ground_truth.split() | |
| hyp_words = hypothesis.split() | |
| gt_index = 0 | |
| hyp_index = 0 | |
| for op in measures['ops']: | |
| for chunk in op: | |
| if chunk.type == 'equal': | |
| # Add equal words without highlighting | |
| highlighted_hyp.extend(gt_words[gt_index:gt_index + (chunk.ref_end_idx - chunk.ref_start_idx)]) | |
| gt_index += (chunk.ref_end_idx - chunk.ref_start_idx) | |
| hyp_index += (chunk.hyp_end_idx - chunk.hyp_start_idx) | |
| elif chunk.type == 'insert': | |
| # Highlight inserted words in green | |
| highlighted_hyp.append(f'<span style="color:green;">{hyp_words[hyp_index]}</span>') | |
| hyp_index += 1 | |
| elif chunk.type == 'sub': | |
| # Highlight substituted words in purple | |
| highlighted_hyp.append(f'<span style="color:purple;">{hyp_words[hyp_index]}</span>') | |
| highlighted_hyp.append(f'<span style="color:red; text-decoration:line-through;">{gt_words[gt_index]}</span>') # Strikethrough for substitution | |
| gt_index += 1 # Move in ground truth | |
| hyp_index += 1 # Move in hypothesis | |
| elif chunk.type == 'delete': | |
| # Highlight deleted words in red with strikethrough | |
| highlighted_hyp.append(f'<span style="color:red; text-decoration:line-through;">{gt_words[gt_index]}</span>') | |
| gt_index += 1 # Move in ground truth | |
| # Handle any remaining words in hypothesis as insertions | |
| while hyp_index < len(hyp_words): | |
| highlighted_hyp.append(f'<span style="color:green;">{hyp_words[hyp_index]}</span>') | |
| hyp_index += 1 | |
| # Handle any remaining words in ground truth that were not matched | |
| while gt_index < len(gt_words): | |
| highlighted_hyp.append(f'<span style="color:red; text-decoration:line-through;">{gt_words[gt_index]}</span>') | |
| gt_index += 1 | |
| highlighted_hyp_str = ' '.join(highlighted_hyp) | |
| error_rate = wer(ground_truth, hypothesis) | |
| # Color Legend HTML | |
| legend_html = """ | |
| <div style="margin-top: 10px;"> | |
| <strong>Legend:</strong><br> | |
| <span style="color:green;">Insertion</span>: Words in green<br> | |
| <span style="color:purple;">Substitution</span>: Words in purple<br> | |
| <span style="color:red; text-decoration:line-through;">Deletion</span>: Words in red with strikethrough<br> | |
| </div> | |
| """ | |
| # Combine highlighted output and legend | |
| combined_output = f"{highlighted_hyp_str}<br>{legend_html}" | |
| return combined_output, error_rate, measures['substitutions'], measures['insertions'], measures['deletions'] | |
| # Gradio Interface | |
| interface = gr.Interface( | |
| fn=highlight_errors, | |
| inputs=["text", "text"], | |
| outputs=[ | |
| gr.Markdown(label="Highlighted Transcript with Legend"), | |
| gr.Number(label="Word Error Rate"), | |
| gr.Number(label="Substitutions"), | |
| gr.Number(label="Insertions"), | |
| gr.Number(label="Deletions") | |
| ], | |
| title="WER Calculator with Error Highlighting and Legend" | |
| ) | |
| interface.launch() | |