## Thursday, March 31, 2011

### [Latex] Simultaneous Equations in Latex

Writing Simultaneous Equations in Latex using cases command, by default, it deems the whole group of equations as one equation and only gives one number for the whole set.

If we want to add numbering to each of the equations inside the group, it's a little difficult.

One solution I found on the web is as follows:

1) define a new command:

\newenvironment{ncases}[4]{%

\parbox{0.5\textwidth}{%

\begin{equation*}

#1=

\begin{cases}

#2

\end{cases}

\end{equation*}}

\hfill

\parbox{1cm}{%

\begin{eqnarray}

{}\label{#3}\\

{}\label{#4}

\end{eqnarray}}}

Note: this command only supports two equations, if needed one can extend it easily to the number you want.

2) Use that command instead of the case command:

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\newenvironment{ncases}[4]{%

\parbox{0.5\textwidth}{%

\begin{equation*}

#1=

\begin{cases}

#2

\end{cases}

\end{equation*}}

\hfill

\parbox{1cm}{%

\begin{eqnarray}

{}\label{#3}\\

{}\label{#4}

\end{eqnarray}}}

\begin{document}

\begin{spacing}{1.1}

\begin{ncases}

{|x|}{x, &\text{if $x\ge 0$}\\

-x,&\text{if $x\le 0$}}{pos}{neg}

\end{ncases}

For if $x\ge 0$, then by Equation~\eqref{pos}, we have $|x|=x\ge0$ and

if $x\le0$, then by Equation~\eqref{neg}, we have $|x|=-x\ge0$.

\end{spacing}

\end{document}

## Wednesday, March 30, 2011

### [Book] Springer Handbook of Speech Processing

This book has a quiet complete explanation over various speech related techniques.

## Monday, March 28, 2011

### [Linux] Prevent a background process from being stopped after closing SSH client

Just find that the job could be run in the background on the server without being killed after log off the SSH terminal. Great!

Moreover, it will send you an email after the job is done!

The wonderful command is:

the 'batch' command.

\$ batch > mycommand -x arg1 -y arg2 -z arg3 > ^D

This stuffs it in to the background, and then mails the results to you. It's a part of cron.

## Friday, March 25, 2011

### [CALL] Mispronunciation detection based on cross-language phonological comparisons

This paper presents a method using speech recognition with linguistic constrains to detect the mispronunciations.

Compared with the standard ASR system, which consists of Acoustic Model, Lexicon and Language Model, the system used for mispronunciation detection only modifies the lexicon to include the possible phoneme confusions for recognition.

The phoneme confusions are gained from cross language phonological comparisons by human beings.

Thus the recognized results would possibly have more errors detected, which are interpreted as mispronunciations.

Actually, we can do alignment instead of recognition for mispronunciation detection. As in learning, the text is known to the speakers and speakers are asked to utter the given sentences.

In this paper, the measures they used are:
1) correctness: the percentage of all correctly detected phones;
2) accuracy: taking account of insertion
3) agreement of the system detection results with human judgments.

## Thursday, March 24, 2011

### [Vis&Speech] TIMIT phoneme distribution

I like figures because "a figure is worth 1000 words".

In the illustration below, the distribution of the 39 phoneme on the standard TIMIT corpus are reflected by the size of the circles.

Meanwhile, the positions of each phoneme actually represented some kind of similarity, although it is not verified by linguistic experts, from what I learned in my English classes, some confusing phones are indeed clustered together.

## Tuesday, March 8, 2011

### [HTK] HLDA transform estimation for High dimensional features

For the TANDEM system, we could either train a set of HMM models on the projected posterior features, we could also train them directly on the posteriors and then using HTK to do a HLDA projection to reduce the dimension of the feature.

However, there is a limitation to the dimension of the feature vectors used: the dimension should be smaller than 100.

This is due to the HMath.c file, in which, there are several functions have defined some local array with the fixed length 100.

For my copy of the v3.4.1 HTK, those are:

1424:   float col[100];
1427:   int n,i,j,perm[100];
1477:   double col[100];
1480:   int n,i,j,perm[100];
1506:   double col[100];
1509:   int n,i,perm[100];
1535:   double col[100];
1538:   int n,i,perm[100];

What needs to do is just increase the size of those local arrays.

### [Speech] Lecture Notes for ASR from Theory to Practice

In the slides, it provides a good list of available tools for automatic speech recognition and thoughts for the future of ASR techniques.

## Monday, March 7, 2011

### [Linguistic] Symbols for American English Vowel Sounds

From:
The position mapping of vowels:

The Sun name for the phoneme is the one used in speech recognition.

A typical dialect of American English has about 15 distinctive vowel sounds. Here their symbols are linked to Sun-style .au samples lifted from the

ibiblio (Sunsite) archive (where they are listed without the .au extension).
• The first symbol is the International Phonetic Association (IPA) symbol for the sound. (For the diphthongs, the American style of transcription is to use a -y where the standard IPA uses a 'j'.)
• The second is the Sun name for the phoneme sample (which is in most cases the same as the symbol used by First Byte in Monologue for Windows and its DOS forebears).
• The third symbol is the ipa-ascii symbol (an alphabet for use on Usenet groups and email).
• The fourth column has the symbol that Rsynth displays in its verbose mode.
• The fifth column contains the SAMPA symbol--as you can see, the differences among these alphabets are minor.
• Each row concludes with a key word for the sound.

 IPA S u n IPAascii Rsynth Sampa KeyWord Front Vowels highlow IY i i i beet IH I I I bit EY eI eI e bait EH E e E bet AE & & { at
 IPA S u n IPAascii Rsynth Sampa KeyWord Back Vowels high.low UY u u u boot UH U U U book OW oU oU o boat AO O O O cause AA a/A A A cot 1
 IPA S u n IPAascii Rsynth Sampa KeyWord Central Vowels AX @ @ @ about AH V V V but2
 IPA S u n IPAascii Rsynth Sampa KeyWord Diphthongs AY aI aI aI bite OY OI OI OI boy AW AU aU aU bough

http://faculty.washington.edu/dillon/PhonResources/PhonResources.html#Recog

http://www.ibiblio.org/sounds/phonemes/

symbols for american english vowel sounds.pdf View this on Posterous

vowel mapping.pdf View this on Posterous

## Thursday, March 3, 2011

### [Cocoa] A simple audio recording and playing demo using AudioQueue

AudioRecord.zip View this on Posterous

This code is based on the Sample project SpeakHere of Apple developer site.

## Tuesday, March 1, 2011

### [Conference] Workshops related to Unsupervised learning and transfer learning

NIPS10_TLWS_Huang.pdf (114 KB)

ICML workshop on Transfer Learning:

http://clopinet.com/isabelle/Projects/ICML2011/

There is also a list of past papers for Transfer learning on that page.

# NIPS 2010 Workshop on Transfer Learning Via Rich Generative Models

Two interesting papers on this NIPS workshop:

### [Misc] Mixing Objective-C and C++

To use the C++ code in Objective-C, we can implement C++ class in pure C++. When using those C++ classes in Objective-C, any file having C++ types must have the extension of ".mm". Thus it seems not suitable to include C++ headers in Objective-C headers, although they are all ".h" files.