Upload cal_wer.py with huggingface_hub
Browse files- cal_wer.py +36 -0
cal_wer.py
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json
|
| 2 |
+
|
| 3 |
+
# 你的输入文件路径
|
| 4 |
+
jsonl_path = "/data2/t-yihanwu/projects/Versa/versa/clean_TTS_wer"
|
| 5 |
+
# "results/VCTK_Demand_miipher-joint-GANloss-noisyspk-90ksteps_wer"
|
| 6 |
+
# "results/podcast_humanLabeled_enUS_segments_wav_miipher-joint-GANloss-noisyspk-miipher-frozen-800ksteps_wer"
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
total_insert = 0
|
| 10 |
+
total_delete = 0
|
| 11 |
+
total_replace = 0
|
| 12 |
+
total_ref_words = 0
|
| 13 |
+
|
| 14 |
+
with open(jsonl_path, "r", encoding="utf-8") as f:
|
| 15 |
+
for line in f:
|
| 16 |
+
entry = json.loads(line)
|
| 17 |
+
ins = entry["whisper_wer_insert"]
|
| 18 |
+
dele = entry["whisper_wer_delete"]
|
| 19 |
+
repl = entry["whisper_wer_replace"]
|
| 20 |
+
equal = entry["whisper_wer_equal"]
|
| 21 |
+
|
| 22 |
+
total_insert += ins
|
| 23 |
+
total_delete += dele
|
| 24 |
+
total_replace += repl
|
| 25 |
+
total_ref_words += (ins + dele + repl + equal)
|
| 26 |
+
|
| 27 |
+
# 避免除以零
|
| 28 |
+
if total_ref_words == 0:
|
| 29 |
+
print("No words in reference. Cannot compute WER.")
|
| 30 |
+
else:
|
| 31 |
+
wer = (total_insert + total_delete + total_replace) / total_ref_words
|
| 32 |
+
print(f"Total Insertions: {total_insert / total_ref_words}")
|
| 33 |
+
print(f"Total Deletions: {total_delete/ total_ref_words}")
|
| 34 |
+
print(f"Total Replacements: {total_replace/ total_ref_words}")
|
| 35 |
+
print(f"Total Reference Words: {total_ref_words/ total_ref_words}")
|
| 36 |
+
print(f"Total WER: {wer:.4f}")
|