From d194b4bcd497838085529376e2dda2825b9db5ee Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期五, 31 三月 2023 15:06:38 +0800
Subject: [PATCH] Merge branch 'dev_cmz2' of github.com:alibaba-damo-academy/FunASR into dev_cmz2 add
---
funasr/export/models/vad_realtime_transformer.py | 28 ++++++++++++++++------------
1 files changed, 16 insertions(+), 12 deletions(-)
diff --git a/funasr/export/models/vad_realtime_transformer.py b/funasr/export/models/vad_realtime_transformer.py
index de7c721..693b9c8 100644
--- a/funasr/export/models/vad_realtime_transformer.py
+++ b/funasr/export/models/vad_realtime_transformer.py
@@ -1,17 +1,12 @@
-from typing import Any
-from typing import List
from typing import Tuple
import torch
import torch.nn as nn
-from funasr.modules.embedding import SinusoidalPositionEncoder
-from funasr.punctuation.sanm_encoder import SANMVadEncoder as Encoder
-from funasr.punctuation.abs_model import AbsPunctuation
-from funasr.punctuation.sanm_encoder import SANMVadEncoder
+from funasr.models.encoder.sanm_encoder import SANMVadEncoder
from funasr.export.models.encoder.sanm_encoder import SANMVadEncoder as SANMVadEncoder_export
-class VadRealtimeTransformer(AbsPunctuation):
+class VadRealtimeTransformer(nn.Module):
def __init__(
self,
@@ -36,8 +31,11 @@
- def forward(self, input: torch.Tensor, text_lengths: torch.Tensor,
- vad_indexes: torch.Tensor) -> Tuple[torch.Tensor, None]:
+ def forward(self, input: torch.Tensor,
+ text_lengths: torch.Tensor,
+ vad_indexes: torch.Tensor,
+ sub_masks: torch.Tensor,
+ ) -> Tuple[torch.Tensor, None]:
"""Compute loss value from buffer sequences.
Args:
@@ -47,7 +45,7 @@
"""
x = self.embed(input)
# mask = self._target_mask(input)
- h, _ = self.encoder(x, text_lengths, vad_indexes)
+ h, _ = self.encoder(x, text_lengths, vad_indexes, sub_masks)
y = self.decoder(h)
return y
@@ -59,10 +57,12 @@
text_indexes = torch.randint(0, self.embed.num_embeddings, (1, length))
text_lengths = torch.tensor([length], dtype=torch.int32)
vad_mask = torch.ones(length, length, dtype=torch.float32)[None, None, :, :]
- return (text_indexes, text_lengths, vad_mask)
+ sub_masks = torch.ones(length, length, dtype=torch.float32)
+ sub_masks = torch.tril(sub_masks).type(torch.float32)
+ return (text_indexes, text_lengths, vad_mask, sub_masks[None, None, :, :])
def get_input_names(self):
- return ['input', 'text_lengths', 'vad_mask']
+ return ['input', 'text_lengths', 'vad_mask', 'sub_masks']
def get_output_names(self):
return ['logits']
@@ -76,6 +76,10 @@
2: 'feats_length1',
3: 'feats_length2'
},
+ 'sub_masks': {
+ 2: 'feats_length1',
+ 3: 'feats_length2'
+ },
'logits': {
1: 'logits_length'
},
--
Gitblit v1.9.1