From 9befa9e508d5ca95cb5faa29cd20d23e04e525c9 Mon Sep 17 00:00:00 2001
From: jmwang66 <wangjiaming.wjm@alibaba-inc.com>
Date: 星期一, 06 二月 2023 16:42:33 +0800
Subject: [PATCH] update data2vec pretrain: add clipping
---
funasr/bin/asr_inference_paraformer.py | 8 +++++---
1 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/funasr/bin/asr_inference_paraformer.py b/funasr/bin/asr_inference_paraformer.py
index 01237f7..de45a33 100644
--- a/funasr/bin/asr_inference_paraformer.py
+++ b/funasr/bin/asr_inference_paraformer.py
@@ -181,7 +181,7 @@
self.nbest = nbest
self.frontend = frontend
self.encoder_downsampling_factor = 1
- if asr_train_args.encoder_conf["input_layer"] == "conv2d":
+ if asr_train_args.encoder == "data2vec_encoder" or asr_train_args.encoder_conf["input_layer"] == "conv2d":
self.encoder_downsampling_factor = 4
@torch.no_grad()
@@ -227,6 +227,8 @@
pre_acoustic_embeds, pre_token_length, alphas, pre_peak_index = predictor_outs[0], predictor_outs[1], \
predictor_outs[2], predictor_outs[3]
pre_token_length = pre_token_length.round().long()
+ if torch.max(pre_token_length) < 1:
+ return []
decoder_outs = self.asr_model.cal_decoder_with_predictor(enc, enc_len, pre_acoustic_embeds, pre_token_length)
decoder_out, ys_pad_lens = decoder_outs[0], decoder_outs[1]
@@ -394,7 +396,7 @@
# results = speech2text(**batch)
# if len(results) < 1:
# hyp = Hypothesis(score=0.0, scores={}, states={}, yseq=[])
-# results = [[" ", ["<space>"], [2], hyp, 10, 6]] * nbest
+# results = [[" ", ["sil"], [2], hyp, 10, 6]] * nbest
# time_end = time.time()
# forward_time = time_end - time_beg
# lfr_factor = results[0][-1]
@@ -621,7 +623,7 @@
results = speech2text(**batch)
if len(results) < 1:
hyp = Hypothesis(score=0.0, scores={}, states={}, yseq=[])
- results = [[" ", ["<space>"], [2], hyp, 10, 6]] * nbest
+ results = [[" ", ["sil"], [2], hyp, 10, 6]] * nbest
time_end = time.time()
forward_time = time_end - time_beg
lfr_factor = results[0][-1]
--
Gitblit v1.9.1