From 8cfee2db5cf7a32f8865f393184d8a48dd6bd38d Mon Sep 17 00:00:00 2001
From: 语帆 <yf352572@alibaba-inc.com>
Date: 星期四, 22 二月 2024 16:22:46 +0800
Subject: [PATCH] test
---
funasr/models/lcbnet/model.py | 46 ++++++++++++++++++++++++++++++++++++++++------
1 files changed, 40 insertions(+), 6 deletions(-)
diff --git a/funasr/models/lcbnet/model.py b/funasr/models/lcbnet/model.py
index c68ccd7..563ff26 100644
--- a/funasr/models/lcbnet/model.py
+++ b/funasr/models/lcbnet/model.py
@@ -1,3 +1,8 @@
+#!/usr/bin/env python3
+# -*- encoding: utf-8 -*-
+# Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
+# MIT License (https://opensource.org/licenses/MIT)
+
import logging
from typing import Union, Dict, List, Tuple, Optional
@@ -16,11 +21,14 @@
from funasr.utils import postprocess_utils
from funasr.utils.datadir_writer import DatadirWriter
from funasr.register import tables
-
-@tables.register("model_classes", "Transformer")
-class Transformer(nn.Module):
- """CTC-attention hybrid Encoder-Decoder model"""
-
+import pdb
+@tables.register("model_classes", "LCBNet")
+class LCBNet(nn.Module):
+ """
+ Author: Speech Lab of DAMO Academy, Alibaba Group
+ LCB-NET: LONG-CONTEXT BIASING FOR AUDIO-VISUAL SPEECH RECOGNITION
+ https://arxiv.org/abs/2401.06390
+ """
def __init__(
self,
@@ -32,10 +40,19 @@
encoder_conf: dict = None,
decoder: str = None,
decoder_conf: dict = None,
+ text_encoder: str = None,
+ text_encoder_conf: dict = None,
+ bias_predictor: str = None,
+ bias_predictor_conf: dict = None,
+ fusion_encoder: str = None,
+ fusion_encoder_conf: dict = None,
ctc: str = None,
ctc_conf: dict = None,
ctc_weight: float = 0.5,
interctc_weight: float = 0.0,
+ select_num: int = 2,
+ select_length: int = 3,
+ insert_blank: bool = True,
input_size: int = 80,
vocab_size: int = -1,
ignore_id: int = -1,
@@ -66,6 +83,15 @@
encoder_class = tables.encoder_classes.get(encoder)
encoder = encoder_class(input_size=input_size, **encoder_conf)
encoder_output_size = encoder.output_size()
+
+ # lcbnet modules: text encoder, fusion encoder and bias predictor
+ text_encoder_class = tables.encoder_classes.get(text_encoder)
+ text_encoder = text_encoder_class(input_size=vocab_size, **text_encoder_conf)
+ fusion_encoder_class = tables.encoder_classes.get(fusion_encoder)
+ fusion_encoder = fusion_encoder_class(**fusion_encoder_conf)
+ bias_predictor_class = tables.encoder_classes.get_class(bias_predictor)
+ bias_predictor = bias_predictor_class(bias_predictor_conf)
+
if decoder is not None:
decoder_class = tables.decoder_classes.get(decoder)
decoder = decoder_class(
@@ -91,6 +117,13 @@
self.specaug = specaug
self.normalize = normalize
self.encoder = encoder
+ # lcbnet
+ self.text_encoder = text_encoder
+ self.fusion_encoder = fusion_encoder
+ self.bias_predictor = bias_predictor
+ self.select_num = select_num
+ self.select_length = select_length
+ self.insert_blank = insert_blank
if not hasattr(self.encoder, "interctc_use_conditioning"):
self.encoder.interctc_use_conditioning = False
@@ -380,7 +413,8 @@
logging.info("enable beam_search")
self.init_beam_search(**kwargs)
self.nbest = kwargs.get("nbest", 1)
-
+ pdb.set_trace()
+
meta_data = {}
if isinstance(data_in, torch.Tensor) and kwargs.get("data_type", "sound") == "fbank": # fbank
speech, speech_lengths = data_in, data_lengths
--
Gitblit v1.9.1