fix misalignment diplay issue

app.py

@@ -33,36 +33,39 @@ def calculate_cer(reference, hypothesis):
 def calculate_sentence_metrics(reference, hypothesis):
     """
     Calculate WER and CER for each sentence and overall statistics.
+    Handles cases where the number of sentences differ.
     """
     try:
         reference_sentences = split_into_sentences(reference)
         hypothesis_sentences = split_into_sentences(hypothesis)
 
-        if len(reference_sentences) != len(hypothesis_sentences):
-            raise ValueError("Reference and hypothesis must contain the same number of sentences")
-
         sentence_wers = []
         sentence_cers = []
-        for ref, hyp in zip(reference_sentences, hypothesis_sentences):
+        min_length = min(len(reference_sentences), len(hypothesis_sentences))
+
+        for i in range(min_length):
+            ref = reference_sentences[i]
+            hyp = hypothesis_sentences[i]
+
             wer = jiwer.wer(ref, hyp)
             cer = jiwer.cer(ref, hyp)
             sentence_wers.append(wer)
             sentence_cers.append(cer)
 
-        if not sentence_wers or not sentence_cers:
-            return {
-                "sentence_wers": [],
-                "sentence_cers": [],
-                "average_wer": 0.0,
-                "average_cer": 0.0,
-                "std_dev_wer": 0.0,
-                "std_dev_cer": 0.0
-            }
-
-        average_wer = np.mean(sentence_wers)
-        average_cer = np.mean(sentence_cers)
-        std_dev_wer = np.std(sentence_wers)
-        std_dev_cer = np.std(sentence_cers)
+        # Calculate overall statistics
+        if sentence_wers:
+            average_wer = np.mean(sentence_wers)
+            std_dev_wer = np.std(sentence_wers)
+        else:
+            average_wer = 0.0
+            std_dev_wer = 0.0
+
+        if sentence_cers:
+            average_cer = np.mean(sentence_cers)
+            std_dev_cer = np.std(sentence_cers)
+        else:
+            average_cer = 0.0
+            std_dev_cer = 0.0
 
         return {
             "sentence_wers": sentence_wers,
@@ -74,19 +77,26 @@ def calculate_sentence_metrics(reference, hypothesis):
         }
     except Exception as e:
         raise e
+
         
 def identify_misaligned_sentences(reference_text, hypothesis_text):
     """
     Identify sentences that don't match between reference and hypothesis.
+    Handles cases where the number of sentences differ.
     Returns a dictionary with misaligned sentence pairs, their indices, and misalignment details.
     """
     reference_sentences = split_into_sentences(reference_text)
     hypothesis_sentences = split_into_sentences(hypothesis_text)
 
     misaligned = []
-    for i, (ref, hyp) in enumerate(zip(reference_sentences, hypothesis_sentences)):
+    min_length = min(len(reference_sentences), len(hypothesis_sentences))
+
+    # Compare sentences up to the minimum length
+    for i in range(min_length):
+        ref = reference_sentences[i]
+        hyp = hypothesis_sentences[i]
+
         if ref != hyp:
-            print(f"Debug: Found misalignment in sentence {i+1}")
             # Find the first position where the sentences diverge
             min_len = min(len(ref), len(hyp))
             misalignment_start = 0
@@ -106,7 +116,29 @@ def identify_misaligned_sentences(reference_text, hypothesis_text):
                 "context_ref": context_ref,
                 "context_hyp": context_hyp
             })
-    print(f"Debug: Total misaligned sentences found: {len(misaligned)}")
+
+    # Note any extra sentences as misaligned
+    if len(reference_sentences) > len(hypothesis_sentences):
+        for i in range(min_length, len(reference_sentences)):
+            misaligned.append({
+                "index": i+1,
+                "reference": reference_sentences[i],
+                "hypothesis": "No corresponding sentence",
+                "misalignment_start": 0,
+                "context_ref": reference_sentences[i],
+                "context_hyp": "No corresponding sentence"
+            })
+    elif len(hypothesis_sentences) > len(reference_sentences):
+        for i in range(min_length, len(hypothesis_sentences)):
+            misaligned.append({
+                "index": i+1,
+                "reference": "No corresponding sentence",
+                "hypothesis": hypothesis_sentences[i],
+                "misalignment_start": 0,
+                "context_ref": "No corresponding sentence",
+                "context_hyp": hypothesis_sentences[i]
+            })
+
     return misaligned
 
 def format_sentence_metrics(sentence_wers, sentence_cers, average_wer, average_cer, std_dev_wer, std_dev_cer, misaligned_sentences):
@@ -130,8 +162,8 @@ def format_sentence_metrics(sentence_wers, sentence_cers, average_wer, average_c
         md += "\n### Misaligned Sentences\n\n"
         for misaligned in misaligned_sentences:
             md += f"#### Sentence {misaligned['index']}\n"
-            md += f"* Reference: {misaligned['reference']}\n"
-            md += f"* Hypothesis: {misaligned['hypothesis']}\n"
+            md += f"* Reference: {misaligned['context_ref']}\n"
+            md += f"* Hypothesis: {misaligned['context_hyp']}\n"
             md += f"* Misalignment starts at position: {misaligned['misalignment_start']}\n\n"
     else:
         md += "\n### Misaligned Sentences\n\n"
@@ -139,7 +171,6 @@ def format_sentence_metrics(sentence_wers, sentence_cers, average_wer, average_c
 
     return md
 
-
 @spaces.GPU()
 def process_files(reference_file, hypothesis_file):
     try:
@@ -168,6 +199,41 @@ def process_files(reference_file, hypothesis_file):
     except Exception as e:
         return {"error": str(e)}
 
+def process_and_display(ref_file, hyp_file):
+    result = process_files(ref_file, hyp_file)
+
+    if "error" in result:
+        error_msg = result["error"]
+        return {"error": error_msg}, "", ""
+
+    metrics = {
+        "Overall WER": result["Overall WER"],
+        "Overall CER": result["Overall CER"]
+    }
+
+    metrics_md = format_sentence_metrics(
+        result["Sentence WERs"],
+        result["Sentence CERs"],
+        result["Average WER"],
+        result["Average CER"],
+        result["Standard Deviation WER"],
+        result["Standard Deviation CER"],
+        result["Misaligned Sentences"]
+    )
+
+    misaligned_md = "### Misaligned Sentences\n\n"
+    if result["Misaligned Sentences"]:
+        for misaligned in result["Misaligned Sentences"]:
+            misaligned_md += f"#### Sentence {misaligned['index']}\n"
+            misaligned_md += f"* Reference: {misaligned['context_ref']}\n"
+            misaligned_md += f"* Hypothesis: {misaligned['context_hyp']}\n"
+            misaligned_md += f"* Misalignment starts at position: {misaligned['misalignment_start']}\n\n"
+    else:
+        misaligned_md += "* No misaligned sentences found."
+
+    return metrics, metrics_md, misaligned_md
+
+
 def main():
     with gr.Blocks() as demo:
         gr.Markdown("# ASR Metrics")
@@ -193,10 +259,10 @@ def main():
 
             if ref_file:
                 with open(ref_file.name, 'r') as f:
-                    ref_text = f.read()[:200]  # Show first 200 characters
+                    ref_text = f.read()[:200]
             if hyp_file:
                 with open(hyp_file.name, 'r') as f:
-                    hyp_text = f.read()[:200]  # Show first 200 characters
+                    hyp_text = f.read()[:200]
 
             return ref_text, hyp_text
 
@@ -211,36 +277,6 @@ def main():
             outputs=[reference_preview, hypothesis_preview]
         )
 
-        def process_and_display(ref_file, hyp_file):
-            result = process_files(ref_file, hyp_file)
-            if "error" in result:
-                error_msg = result["error"]
-                return {"error": error_msg}, "", "", {"error": error_msg}
-
-            metrics = {
-                "Overall WER": result["Overall WER"],
-                "Overall CER": result["Overall CER"]
-            }
-
-            metrics_md = format_sentence_metrics(
-                result["Sentence WERs"],
-                result["Sentence CERs"],
-                result["Average WER"],
-                result["Average CER"],
-                result["Standard Deviation WER"],
-                result["Standard Deviation CER"],
-                result["Misaligned Sentences"]
-            )
-
-            misaligned_md = "### Misaligned Sentences\n\n"
-            for misaligned in result["Misaligned Sentences"]:
-                misaligned_md += f"#### Sentence {misaligned['index']}\n"
-                misaligned_md += f"* Reference: {misaligned['context_ref']}\n"
-                misaligned_md += f"* Hypothesis: {misaligned['context_hyp']}\n"
-                misaligned_md += f"* Misalignment starts at position: {misaligned['misalignment_start']}\n\n"
-
-            return metrics, metrics_md, misaligned_md
-
         compute_button.click(
             fn=process_and_display,
             inputs=[reference_file, hypothesis_file],