jasonyao81000/dsp2018spring

Fundamentals of Speech Signal Processing at NTU 2018 Spring.

DSP2018SPRING

Fundamentals of Speech Signal Processing at NTU 2018 Spring.

This course has three homeworks.

Table of Contents

1. Homework 1
2. Homework 2
3. Homework 3

Homework 1

I. ENVIRONMENT

[email protected]

II. HOW TO EXECUTE

2.1. Make

cd hw1_r06922002/
make

2.2. Train

./train 1000 ../model_init.txt ../seq_model_01.txt ./model_01.txt
./train 1000 ../model_init.txt ../seq_model_02.txt ./model_02.txt
./train 1000 ../model_init.txt ../seq_model_03.txt ./model_03.txt
./train 1000 ../model_init.txt ../seq_model_04.txt ./model_04.txt
./train 1000 ../model_init.txt ../seq_model_05.txt ./model_05.txt

2.3. Test

./test ../modellist.txt ../testing_data1.txt ../result1.txt
./test ../modellist.txt ../testing_data2.txt ../result2.txt

III. RESULTS

---

Homework 2

I. ENVIRONMENT

[email protected]

II. HOW TO EXECUTE

cd hw2_r06922002/
bash 00_clean_all.sh
bash 01_run_HCopy.sh
bash 02_run_HCompV.sh
bash 03_training.sh
bash 04_testing.sh
cat result/accuracy

III. Run Baseline (40%)

IV. Imporve Accuracy (40%)

---

Homework 3

I. ENVIRONMENT

[email protected]

II. HOW TO COMPILE

cd hw3_r06922002/
make clean
copy TA’s bigram.lm, Big5-ZhuYin.map, testdata to ./hw3_r06922002/
make MACHINE_TYPE=i686-m64 SRIPATH=/home/master/06/r06922002/DSP2018Spring/srilm-1.5.10 all

III. HOW TO EXECUTE

cd hw3_r06922002/
make clean
copy TA’s bigram.lm, Big5-ZhuYin.map, testdata to ./r06922002/
make MACHINE_TYPE=i686-m64 SRIPATH=/home/master/06/r06922002/DSP2018Spring/srilm-1.5.10 all
make map
make MACHINE_TYPE=i686-m64 SRIPATH=/home/master/06/r06922002/DSP2018Spring/srilm-1.5.10 run

IV. WHAT I HAVE DONE

4.1. Segment corpus and all test data into characters

./separator_big5.pl corpus.txt > corpus_seg.txt
./separator_big5.pl testdata/xx.txt > xx.txt

4.2. Train character-based bigram LM (Bigram)

#!/bin/bash
SRIPATH="/home/master/06/r06922002/DSP2018Spring/srilm-1.5.10"
SRIPATH_BIN="$SRIPATH/bin/i686-m64"
$SRIPATH_BIN/ngram-count -text corpus_seg.txt -write lm.cnt -order 2
$SRIPATH_BIN/ngram-count -read lm.cnt -lm bigram.lm -unk -order 2

4.3. Generate ZhuYin-Big5.map from Big5-ZhuYin.map

python mapping.py Big5-ZhuYin.map ZhuYin-Big5.map

or

make map

4.4. Using disambig to decode testdata/xx.txt (Bigram)

#!/bin/bash
SRIPATH="/home/master/06/r06922002/DSP2018Spring/srilm-1.5.10"
SRIPATH_BIN="$SRIPATH/bin/i686-m64"
$SRIPATH_BIN/disambig -text testdata/xx.txt -map ZhuYin-Big5.map -lm bigram.lm -order 2 > result1/xx.txt

4.5. Using mydisambig to decode testdata/xx.txt (Bigram)

./mydisambig -text testdata/xx.txt -map ZhuYin-Big5.map -lm bigram.lm -order 2 > result2/xx.txt

or

make run

4.6. Results (Bigram)