Skip to content

Instantly share code, notes, and snippets.

View SnowMasaya's full-sized avatar

SnowGushiGit SnowMasaya

36 followers · 6 following
View GitHub Profile
@SnowMasaya
SnowMasaya / file0.txt
Created April 10, 2015 01:57
音声認識システム　Kaldiを試しに動かしてみた ref: http://qiita.com/GushiSnow/items/43d5916cc8a0c939f1dd
build-essential
gfortran
libgfortran3
python-dev(python3-dev)
libblas-dev
libatlas-base-dev
cython
g++
zlib1g-dev
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:19
Kaldiに関する処理を日本語のドキュメントでまとめてみた(データ準備編）１ ref: http://qiita.com/GushiSnow/items/cc1440e0a8ea199e78c5
getdata.sh
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:19
Kaldiに関する処理を日本語のドキュメントでまとめてみた（データ準備編）2 ref: http://qiita.com/GushiSnow/items/a24cad7231de341738ee
#silのみ出力
utils/make_lexicon_fst_silprob.pl $tmpdir/lexiconp_silprob_disambig.txt $s rcdir/silprob.txt $silphone '#'$ndisambig | \
#置き換え処理
sed 's=\#[0-9][0-9]*=<eps>=g' | \for indirect one, use twice the learning rate
#音素を入力、単語を出力として重み付き状態変換器の作成
fstcompile --isymbols=$dir/phones.txt --osymbols=$dir/words.txt \
--keep_isymbols=false --keep_osymbols=false | \
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:19
Kaldiに関する処理を日本語のドキュメントでまとめてみた（特徴量抽出編）3 ref: http://qiita.com/GushiSnow/items/e099baf9d1c2e72cb3d1
for x in train test; do
steps/make_mfcc.sh --cmd "$train_cmd" --nj $njobs \
data/$x exp/make_mfcc/$x $mfccdir || exit 1;
steps/compute_cmvn_stats.sh data/$x exp/make_mfcc/$x $mfccdir || exit 1;
done
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:19
Kaldiに関する処理を日本語のドキュメントでまとめてみた（学習編）4 ref: http://qiita.com/GushiSnow/items/d431a5c49dc4206def2d
utils/subset_data_dir.sh data/train 1000 data/train.1k || exit 1;
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:19
Kaldiに関する処理を日本語のドキュメントでまとめてみた（グラフ作成編）5 ref: http://qiita.com/GushiSnow/items/8e1c25b1d2eda8c1f2c3
"--mono" #モノフォン
"--quinphone" #クインフォン
"--reverse" #逆順
"--transition-scale" #スケール変換
"--self-loop-scale" #自己学習の回数スケール
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:19
Kaldiに関する処理を日本語のドキュメントでまとめてみた（デコーディング編）6 ref: http://qiita.com/GushiSnow/items/01296c16f0d9d823ae55
if [ -z "$model" ]; then # if --model <mdl> was not specified on the command lin
e...
if [ -z $iter ]; then model=$srcdir/final.mdl;
else model=$srcdir/$iter.mdl; fi
fi
for f in $sdata/1/feats.scp $sdata/1/cmvn.scp $model $graphdir/HCLG.fst; do
[ ! -f $f ] && echo "decode.sh: no such file $f" && exit 1;
done
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:19
音声対話システムの評価方法 ref: http://qiita.com/GushiSnow/items/2dd028510de1203c857c
\begin{aligned}
P(E) & = \sum_i^n (\frac{t_i}{T})^2 \\
\end{aligned}
@SnowMasaya
SnowMasaya / file0.py
Last active August 29, 2015 14:21
PyBrainを用いて強化学習をしてみた ref: http://qiita.com/GushiSnow/items/bfb51afc7d57451bc036
from scipy import *
import sys, time
from pybrain.rl.environments.mazes import Maze, MDPMazeTask
from pybrain.rl.learners.valuebased import ActionValueTable
from pybrain.rl.agents import LearningAgent
from pybrain.rl.learners import Q, SARSA
from pybrain.rl.experiments import Experiment
from pybrain.rl.environments import Task
@SnowMasaya
SnowMasaya / file0.txt
Last active August 29, 2015 14:21
The Unreasonable Effectiveness of Recurrent Neural Networks　文字ごとのLSTMニューラルネット言語モデルでテキスト(シェークスピア)、Wikipediaページ、コード(Linuxのカーネル)などを自動生成する話 ref: http://qiita.com/GushiSnow/items/c0ff3f213cade8b760b8
rnn = RNN()
y = rnn.step(x) # x is an input vector, y is the RNN's output vector