From 8c6d1642f5fbf1d55edb324e35e9fa6e89da25a1 Mon Sep 17 00:00:00 2001
From: Xuning Tan <me@xuning.eu.org>
Date: 星期三, 01 十月 2025 14:45:45 +0800
Subject: [PATCH] fix: correct the deepspeed_config path reference in the finetuning script (#2680)
---
funasr/models/sense_voice/model.py | 18 ++++++++++++++----
1 files changed, 14 insertions(+), 4 deletions(-)
diff --git a/funasr/models/sense_voice/model.py b/funasr/models/sense_voice/model.py
index bd1c76d..6a29181 100644
--- a/funasr/models/sense_voice/model.py
+++ b/funasr/models/sense_voice/model.py
@@ -925,6 +925,11 @@
if tok_ls: token_ids.extend(tok_ls)
else: token_ids.append(124)
+ if len(token_ids) == 0:
+ result_i = {"key": key[i], "text": text}
+ results.append(result_i)
+ continue
+
logits_speech = self.ctc.softmax(encoder_out)[i, 4 : encoder_out_lens[i].item(), :]
pred = logits_speech.argmax(-1).cpu()
logits_speech[pred == self.blank_id, self.blank_id] = 0
@@ -947,8 +952,8 @@
timestamp.append([tokens[token_id], ts_left, ts_right])
token_id += 1
_start = _end
- timestamp = self.post(timestamp)
- result_i = {"key": key[i], "text": text, "timestamp": timestamp}
+ timestamp, words = self.post(timestamp)
+ result_i = {"key": key[i], "text": text, "timestamp": timestamp, "words": words}
results.append(result_i)
else:
result_i = {"key": key[i], "text": text}
@@ -957,6 +962,7 @@
def post(self, timestamp):
timestamp_new = []
+ words_new = []
prev_word = None
for i, t in enumerate(timestamp):
word, start, end = t
@@ -967,17 +973,21 @@
if i == 0:
# timestamp_new.append([word, start, end])
timestamp_new.append([start, end])
+ words_new.append(word)
elif word.startswith("鈻�"):
word = word[1:]
timestamp_new.append([start, end])
+ words_new.append(word)
elif prev_word is not None and prev_word.isalpha() and prev_word.isascii() and word.isalpha() and word.isascii():
- prev_word += word
+ word = prev_word + word
timestamp_new[-1][1] = end
+ words_new[-1] = word
else:
# timestamp_new[-1][0] += word
timestamp_new.append([start, end])
+ words_new.append(word)
prev_word = word
- return timestamp_new
+ return timestamp_new, words_new
def export(self, **kwargs):
from .export_meta import export_rebuild_model
--
Gitblit v1.9.1