From b5ea9c7a6a7cf0816fd59d7b3377752390d3a775 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E6=B8=B8=E9=9B=81?= <zhifu.gzf@alibaba-inc.com>
Date: Thu, 29 Feb 2024 14:33:10 +0800
Subject: [PATCH] cer

---
 funasr/metrics/{compute_wer.py => wer.py} | 42 +++++++++++++++++------
 funasr/models/llm_asr_nar/model.py        |  3 +-
 2 files changed, 34 insertions(+), 11 deletions(-)
 rename funasr/metrics/{compute_wer.py => wer.py} (81%)

diff --git a/funasr/metrics/compute_wer.py b/funasr/metrics/wer.py
similarity index 81%
rename from funasr/metrics/compute_wer.py
rename to funasr/metrics/wer.py
index 26a9f491f..77abbbde7 100755
--- a/funasr/metrics/compute_wer.py
+++ b/funasr/metrics/wer.py
@@ -1,10 +1,13 @@
 import os
 import numpy as np
 import sys
+import hydra
 
 def compute_wer(ref_file,
                 hyp_file,
-                cer_detail_file):
+                cer_file,
+                cn_postprocess=False,
+                ):
     rst = {
         'Wrd': 0,
         'Corr': 0,
@@ -24,14 +27,22 @@ def compute_wer(ref_file,
         for line in hyp_reader:
             key = line.strip().split()[0]
             value = line.strip().split()[1:]
+            if cn_postprocess:
+                value = value.replace(" ", "")
+                value = [x for x in value]
+                value = " ".join(value)
             hyp_dict[key] = value
     with open(ref_file, 'r') as ref_reader:
         for line in ref_reader:
             key = line.strip().split()[0]
             value = line.strip().split()[1:]
+            if cn_postprocess:
+                value = value.replace(" ", "")
+                value = [x for x in value]
+                value = " ".join(value)
             ref_dict[key] = value
 
-    cer_detail_writer = open(cer_detail_file, 'w')
+    cer_detail_writer = open(cer_file, 'w')
     for hyp_key in hyp_dict:
         if hyp_key in ref_dict:
            out_item = compute_wer_by_line(hyp_dict[hyp_key], ref_dict[hyp_key])
@@ -47,6 +58,7 @@ def compute_wer(ref_file,
            cer_detail_writer.write(hyp_key + print_cer_detail(out_item) + '\n')
            cer_detail_writer.write("ref:" + '\t' + " ".join(list(map(lambda x: x.lower(), ref_dict[hyp_key]))) + '\n')
            cer_detail_writer.write("hyp:" + '\t' + " ".join(list(map(lambda x: x.lower(), hyp_dict[hyp_key]))) + '\n')
+           cer_detail_writer.flush()
 
     if rst['Wrd'] > 0:
         rst['Err'] = round(rst['wrong_words'] * 100 / rst['Wrd'], 2)
@@ -59,6 +71,7 @@ def compute_wer(ref_file,
     cer_detail_writer.write("%SER " + str(rst['S.Err']) + " [ " + str(rst['wrong_sentences']) + " / " + str(rst['Snt']) + " ]" + '\n')
     cer_detail_writer.write("Scored " + str(len(hyp_dict)) + " sentences, " + str(len(hyp_dict) - rst['Snt']) + " not present in hyp." + '\n')
 
+    cer_detail_writer.close()
      
 def compute_wer_by_line(hyp,
                         ref):
@@ -146,12 +159,21 @@ def print_cer_detail(rst):
             + str(rst['sub']) + ") corr:" + '{:.2%}'.format(rst['cor']/rst['nwords'])
             + ",cer:" + '{:.2%}'.format(rst['wrong']/rst['nwords']))
 
-if __name__ == '__main__':
-    if len(sys.argv) != 4:
-        print("usage : python compute-wer.py test.ref test.hyp test.wer")
-        sys.exit(0)
 
-    ref_file = sys.argv[1]
-    hyp_file = sys.argv[2]
-    cer_detail_file = sys.argv[3]
-    compute_wer(ref_file, hyp_file, cer_detail_file)
+@hydra.main(config_name=None, version_base=None)
+def main_hydra(cfg: DictConfig):
+    ref_file = cfg.get("ref_file", None)
+    hyp_file = cfg.get("hyp_file", None)
+    cer_file = cfg.get("cer_file", None)
+    cn_postprocess = cfg.get("cn_postprocess", False)
+    if ref_file is None or hyp_file is None or cer_file is None:
+        print("usage : python -m  funasr.metrics.wer ++ref_file=test.ref ++hyp_file=test.hyp ++cer_file=test.wer ++cn_postprocess=false")
+        sys.exit(0)
+        
+    compute_wer(ref_file, hyp_file, cer_file, cn_postprocess)
+
+if __name__ == '__main__':
+    main_hydra()
+
+
+
diff --git a/funasr/models/llm_asr_nar/model.py b/funasr/models/llm_asr_nar/model.py
index 6a4eccefe..f17034930 100644
--- a/funasr/models/llm_asr_nar/model.py
+++ b/funasr/models/llm_asr_nar/model.py
@@ -315,7 +315,8 @@ class LLMASRNAR(nn.Module):
         model_outputs = self.llm(inputs_embeds=inputs_embeds, attention_mask=attention_mask, labels=None)
         preds = torch.argmax(model_outputs.logits, -1)
         text = tokenizer.batch_decode(preds, add_special_tokens=False, skip_special_tokens=True)
-        text = text[0].split(': \n')[-1]
+        text = text[0].split(': ')[-1]
+        text = text.strip()
         # preds = torch.argmax(model_outputs.logits, -1)
         
         ibest_writer = None