speech

中文语音识别训练，使用3层cnn+多层bilstm实现。使用lmdb存放音频特征MFCC，调用caffe 的函数在存放到lmdb里面做个矩阵转换。交叉熵损失函数和ctc_loss都做了尝试。训练集为清语音集。一个特征对应一个单字。

Name		Name	Last commit message	Last commit date
Latest commit History 12 Commits
doc		doc
Makefile.config		Makefile.config
README		README
README.md		README.md
build_lmdb.py		build_lmdb.py
build_lmdb2.py		build_lmdb2.py
build_lmdb_step.py		build_lmdb_step.py
build_mfcc.py		build_mfcc.py
build_mfcc2.py		build_mfcc2.py
build_records.py		build_records.py
build_records2.py		build_records2.py
build_records3.py		build_records3.py
bulild_lmdb.py		bulild_lmdb.py
drop_mfcc.py		drop_mfcc.py
filename.pkl		filename.pkl
make_lmdb.py		make_lmdb.py
make_mfcc_sentence.py		make_mfcc_sentence.py
makerecord.py		makerecord.py
new.py		new.py
new_train.py		new_train.py
new_train2.py		new_train2.py
pinyin		pinyin
read_dic.py		read_dic.py
read_mfcc.py		read_mfcc.py
read_pkl.py		read_pkl.py
read_records.py		read_records.py
read_txt.py		read_txt.py
readtfrecord.py		readtfrecord.py
readtxt.py		readtxt.py
readwav.py		readwav.py
sample.py		sample.py
speech_m_gpu5.py		speech_m_gpu5.py
speech_m_gpu6.py		speech_m_gpu6.py
speech_test.py		speech_test.py
speech_train.py		speech_train.py
speech_train2.py		speech_train2.py
speech_train_3.py		speech_train_3.py
speech_train_new.py		speech_train_new.py
speech_train_new3.py		speech_train_new3.py
speech_train_new5.py		speech_train_new5.py
ssd.py		ssd.py
windows.txt		windows.txt

Provide feedback