add cer

app.py

@@ -9,9 +9,7 @@ def split_into_sentences(text):
     Simple sentence tokenizer using regular expressions.
     Splits text into sentences based on punctuation.
     """
-    # Split text into sentences using regex
     sentences = re.split(r'(?<=[.!?])\s*', text)
-    # Clean up empty strings and whitespace
     sentences = [s.strip() for s in sentences if s.strip()]
     return sentences
 
@@ -32,9 +30,9 @@ def calculate_cer(reference, hypothesis):
     return cer
 
 @spaces.GPU()
-def calculate_sentence_wer(reference, hypothesis):
+def calculate_sentence_metrics(reference, hypothesis):
     """
-    Calculate WER for each sentence and overall statistics.
+    Calculate WER and CER for each sentence and overall statistics.
     """
     try:
         reference_sentences = split_into_sentences(reference)
@@ -44,24 +42,35 @@ def calculate_sentence_wer(reference, hypothesis):
             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):
-            sentence_wer = jiwer.wer(ref, hyp)
-            sentence_wers.append(sentence_wer)
+            wer = jiwer.wer(ref, hyp)
+            cer = jiwer.cer(ref, hyp)
+            sentence_wers.append(wer)
+            sentence_cers.append(cer)
 
-        if not sentence_wers:
+        if not sentence_wers or not sentence_cers:
             return {
                 "sentence_wers": [],
+                "sentence_cers": [],
                 "average_wer": 0.0,
-                "std_dev": 0.0
+                "average_cer": 0.0,
+                "std_dev_wer": 0.0,
+                "std_dev_cer": 0.0
             }
 
         average_wer = np.mean(sentence_wers)
-        std_dev = np.std(sentence_wers)
+        average_cer = np.mean(sentence_cers)
+        std_dev_wer = np.std(sentence_wers)
+        std_dev_cer = np.std(sentence_cers)
 
         return {
             "sentence_wers": sentence_wers,
+            "sentence_cers": sentence_cers,
             "average_wer": average_wer,
-            "std_dev": std_dev
+            "average_cer": average_cer,
+            "std_dev_wer": std_dev_wer,
+            "std_dev_cer": std_dev_cer
         }
     except Exception as e:
         raise e
@@ -75,30 +84,43 @@ def process_files(reference_file, hypothesis_file):
         with open(hypothesis_file.name, 'r') as f:
             hypothesis_text = f.read()
 
-        wer_value = calculate_wer(reference_text, hypothesis_text)
-        cer_value = calculate_cer(reference_text, hypothesis_text)
-        sentence_wer_stats = calculate_sentence_wer(reference_text, hypothesis_text)
+        overall_wer = calculate_wer(reference_text, hypothesis_text)
+        overall_cer = calculate_cer(reference_text, hypothesis_text)
+        sentence_metrics = calculate_sentence_metrics(reference_text, hypothesis_text)
 
         return {
-            "WER": wer_value,
-            "CER": cer_value,
-            "Sentence WERs": sentence_wer_stats["sentence_wers"],
-            "Average WER": sentence_wer_stats["average_wer"],
-            "Standard Deviation": sentence_wer_stats["std_dev"]
+            "Overall WER": overall_wer,
+            "Overall CER": overall_cer,
+            "Sentence WERs": sentence_metrics["sentence_wers"],
+            "Sentence CERs": sentence_metrics["sentence_cers"],
+            "Average WER": sentence_metrics["average_wer"],
+            "Average CER": sentence_metrics["average_cer"],
+            "Standard Deviation WER": sentence_metrics["std_dev_wer"],
+            "Standard Deviation CER": sentence_metrics["std_dev_cer"]
         }
     except Exception as e:
         return {"error": str(e)}
 
-def format_sentence_wer_stats(sentence_wers, average_wer, std_dev):
-    if not sentence_wers:
+def format_sentence_metrics(sentence_wers, sentence_cers, average_wer, average_cer, std_dev_wer, std_dev_cer):
+    if not sentence_wers and not sentence_cers:
         return "All sentences match perfectly!"
 
-    md = "### Sentence-level WER Analysis\n\n"
+    md = "### Sentence-level Metrics\n\n"
+    md += "#### Word Error Rate (WER)\n"
     md += f"* Average WER: {average_wer:.2f}\n"
-    md += f"* Standard Deviation: {std_dev:.2f}\n\n"
+    md += f"* Standard Deviation: {std_dev_wer:.2f}\n\n"
+    md += "#### Character Error Rate (CER)\n"
+    md += f"* Average CER: {average_cer:.2f}\n"
+    md += f"* Standard Deviation: {std_dev_cer:.2f}\n\n"
+
     md += "### WER for Each Sentence\n\n"
     for i, wer in enumerate(sentence_wers):
         md += f"* Sentence {i+1}: {wer:.2f}\n"
+
+    md += "\n### CER for Each Sentence\n\n"
+    for i, cer in enumerate(sentence_cers):
+        md += f"* Sentence {i+1}: {cer:.2f}\n"
+
     return md
 
 def main():
@@ -116,7 +138,7 @@ def main():
         with gr.Row():
             compute_button = gr.Button("Compute Metrics")
             results_output = gr.JSON(label="Results")
-            wer_stats_output = gr.Markdown(label="WER Statistics")
+            metrics_output = gr.Markdown(label="Sentence Metrics")
 
         # Update previews when files are uploaded
         def update_previews(ref_file, hyp_file):
@@ -149,22 +171,25 @@ def main():
                 return {}, {}, "Error: " + result["error"]
 
             metrics = {
-                "WER": result["WER"],
-                "CER": result["CER"]
+                "Overall WER": result["Overall WER"],
+                "Overall CER": result["Overall CER"]
             }
 
-            wer_stats_md = format_sentence_wer_stats(
+            metrics_md = format_sentence_metrics(
                 result["Sentence WERs"],
+                result["Sentence CERs"],
                 result["Average WER"],
-                result["Standard Deviation"]
+                result["Average CER"],
+                result["Standard Deviation WER"],
+                result["Standard Deviation CER"]
             )
 
-            return metrics, wer_stats_md
+            return metrics, metrics_md
 
         compute_button.click(
             fn=process_and_display,
             inputs=[reference_file, hypothesis_file],
-            outputs=[results_output, wer_stats_output]
+            outputs=[results_output, metrics_output]
         )
 
     demo.launch()