mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-09 18:12:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

remove ineffective normalize_text_alimeeting

Browse Source
This commit is contained in:
yfyeung 2025-04-29 06:24:43 -07:00 committed by Your Name
parent 46b9be31cc
commit 26aef4a926
1 changed files with 3 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
egs/speech_llm/ASR_LLM/whisper_llm_zh/decode.py
View File

@ -357,43 +357,6 @@ def decode_dataset(
Returns:
Return a dict, whose key may be "beam-search".
"""
def normalize_text_alimeeting(text: str, normalize: str = "m2met") -> str:
"""
Text normalization similar to M2MeT challenge baseline.
See: https://github.com/yufan-aslp/AliMeeting/blob/main/asr/local/text_normalize.pl
"""
if normalize == "none":
return text
elif normalize == "m2met":
import re
text = text.replace(" ", "")
text = text.replace("<sil>", "")
text = text.replace("<%>", "")
text = text.replace("<->", "")
text = text.replace("<$>", "")
text = text.replace("<#>", "")
text = text.replace("<_>", "")
text = text.replace("<space>", "")
text = text.replace("`", "")
text = text.replace("&", "")
text = text.replace(",", "")
if re.search("[a-zA-Z]", text):
text = text.upper()
text = text.replace("Ａ", "A")
text = text.replace("ａ", "A")
text = text.replace("ｂ", "B")
text = text.replace("ｃ", "C")
text = text.replace("ｋ", "K")
text = text.replace("ｔ", "T")
text = text.replace("，", "")
text = text.replace("丶", "")
text = text.replace("。", "")
text = text.replace("、", "")
text = text.replace("？", "")
return text
results = []
num_cuts = 0
@ -406,6 +369,7 @@ def decode_dataset(
results = defaultdict(list)
for batch_idx, batch in enumerate(dl):
texts = batch["supervisions"]["text"]
texts = [list("".join(text.split())) for text in texts]
cut_ids = [cut.id for cut in batch["supervisions"]["cut"]]
hyps_dict = decode_one_batch(
@ -419,11 +383,9 @@ def decode_dataset(
this_batch = []
assert len(hyps) == len(texts)
for cut_id, hyp_words, ref_text in zip(cut_ids, hyps, texts):
ref_text = normalize_text_alimeeting(ref_text)
ref_words = ref_text.split()
print(f"ref: {ref_text}")
print(f"ref: {''.join(ref_text)}")
print(f"hyp: {''.join(hyp_words)}")
this_batch.append((cut_id, ref_words, hyp_words))
this_batch.append((cut_id, ref_text, hyp_words))
results[lm_scale].extend(this_batch)