From e5dd9cb090701f64e002aed1da7bd1af1039ca64 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期一, 18 三月 2024 21:09:39 +0800
Subject: [PATCH] paraformer streaming bugfix
---
funasr/models/paraformer_streaming/model.py | 12 +++++-------
1 files changed, 5 insertions(+), 7 deletions(-)
diff --git a/funasr/models/paraformer_streaming/model.py b/funasr/models/paraformer_streaming/model.py
index 518fe93..a53aad5 100644
--- a/funasr/models/paraformer_streaming/model.py
+++ b/funasr/models/paraformer_streaming/model.py
@@ -531,6 +531,8 @@
for i in range(n):
kwargs["is_final"] = _is_final and i == n -1
audio_sample_i = audio_sample[i*chunk_stride_samples:(i+1)*chunk_stride_samples]
+ if kwargs["is_final"] and len(audio_sample_i) < 960:
+ continue
# extract fbank feats
speech, speech_lengths = extract_fbank([audio_sample_i], data_type=kwargs.get("data_type", "sound"),
@@ -579,14 +581,10 @@
decoder_class = tables.decoder_classes.get(kwargs["decoder"] + "Export")
self.decoder = decoder_class(self.decoder, onnx=is_onnx)
- from funasr.utils.torch_function import MakePadMask
from funasr.utils.torch_function import sequence_mask
-
- if is_onnx:
- self.make_pad_mask = MakePadMask(max_seq_len, flip=False)
- else:
- self.make_pad_mask = sequence_mask(max_seq_len, flip=False)
-
+
+
+ self.make_pad_mask = sequence_mask(max_seq_len, flip=False)
import copy
import types
--
Gitblit v1.9.1