This is the first part of another question of mine that had a recommendation to make it two questions: Adding Accents to Speech Generation.
Summary: The other question asks how to add an accent programatically to generated speech. Not an accent mark or inflection, but a full accent like a British or Scottish or Russian one.
The first question (same as this one) asks how the original text could be analyzed to determine what accents need to be added and where.
Basically, how could text be analyzed to find these accents and generate a set of instructions that could be used to add any accent to any generated speech?
Related
I got two different "versions" of Arabic letters on Wikipedia. The first example seems to be 3 sub-components in one:
"ـمـ".split('').map(x => x.codePointAt(0).toString(16))
[ '640', '645', '640' ]
Finding this "m medial" letter on this page gives me this:
ﻤ
fee4
The code points 640 and 645 are the "Arabic tatwheel" ـ and "Arabic letter meem" م. What the heck? How does this work? I don't see anywhere in the information so far on Unicode Arabic how these glyphs are "composed". Why is it composed from these parts? Is there a pattern for the structure of all glyphs? (All the glyphs on the first Wikipedia page are similar, but the second they are one code point). Where do I find information on how to parse out the characters effectively in Arabic (or any other language for that matter)?
Arabic is a script with cursive joining; the shape of the letters changes depending on whether they occur initially, medially, or finally within a word. Sometimes you may want to display these contextual forms in isolation, for example to simply show what they look like.
The recommended way to go about this is by using special join-causing characters for the letters to connect to. One of these is the tatweel (also called kashida), which is essentially a short line segment with “glue” at each end. So if you surround the letter م with a tatweel character on both sides, the text renderer automatically selects its medial form as if it occured in the middle of a word (ـمـ). The underlying character code of the م doesn’t change, only its visible glyph.
However, for historical reasons Unicode also contains a large set of so-called presentation forms for Arabic. These represent those same contextual letter shapes, but as separate character codes that do not change depending on their surroundings; putting the “isolated” presentation form of م between two tatweels does not affect its appearance, for instance: ـﻡـ
It is not recommended to use these presentation forms for actually writing Arabic. They exist solely for compatibility with old legacy encodings and aren’t needed for correctly typesetting Arabic text. Wikipedia just used them for demonstration purposes and to show off that they exist, I presume. If you encounter presentation forms, you can usually apply Unicode normalisation (NFKD or NFKC) to the string to get the underlying base letters. See the Unicode FAQ on presentation forms for more information.
I am trying to adapt en-us acoustic model with indian english accent recordings. Since many words are pronounced in different accent, do I need to add the updated phoneme representation of words? Currently I am following this link: https://cmusphinx.github.io/wiki/tutorialadapt/#accumulating-observation-counts and here nothing is mentioned about updating your .dict file.
PS: Should I add new words directly in the dictionary?
There is Indian English model in downloads, you should use it instead. It comes with Indian English dictionary.
I am currently coding in python and managed to use pdftotext in order to extract the text from a pdf.
That particular text file is split up in a list of strings. By using regular expression I am able to find specific words I am interested in. The reason why I divide the text into a list is that I want to measure the distance between two specific words and by distance I mean the number of words in between the two words.
However after finding the position of the words I would like to be able to refer back to the initial pdf. In detail, I am interested in the page and maybe even line (if pdf supports this kind of structure) where these words are located.
One idea I have is to do this process for each page of the pdf, so when I find these words I know on what page this was. But this has the big disadvantage that sometimes page breaks are not necessarily natural. Meaning, I would lose the ability to find the words if they are unfortunately separated by a page break.
Do you have any idea how to do this in a more sophisticated manner?
You'll need a more sophisticated library than the one you're using. The Datalogics PDF Java Toolkit has several classes that can extract text from a PDF file. The one you use depends on what you want to do with the text after extraction. The ReadingOrderTextExtractor will create a list of lists that will allow you to extract the text and examine the content of paragraphs, sentences within those paragraphs, and words within that sentence. You'll not only be able to tell the distance between the words but whether they are in the same sentence or paragraph. One you've found a Word object, you can then find both it's location on the page, allowing for highlighting, and the page number it's on.
I'm searching for a tool that converts text to phonemes, (like text to speech software)
I can program one but it will not be without errors and takes a lot of time!
so my question is:
is there a simple tool for converting e.g.
"hello" to "HH AH0 L OW1"
maybe some command-line tool so i can capture the stdout?
i'm searching for the phonemes in 'Arpabet' style (see the 'hello' example).
espeak does something like that but the output is not in Arpabet style and the phonemes are
not split by some determiner.
If you had searched for Arpabet on wiki you would have found your answer. The CMU guys have prepared scripts which convert most english words to their respective Arpabet phonetic break up.
If you want the phone sequence of a couple of words you can use their interface here. But, if you want it for a big file then you might have to run their scripts on your own. They used to have a working page here, but it seems to be not working now.
What kind of sign is "" and what is it used for (note there is a invisible sign there)?
I have searched through all my documents and found a lot of them. They messed upp my htaccess file. I think I got them when I copied webadresses from google to redirect. So maybe a warning searching through your documents for this one also :)
It is U+200E LEFT-TO-RIGHT MARK. (A quick way to check out such things is to copy a string containing the character and paste it in the writing area in my Full Unicode input utility, then click on the “Show U+” button there, and use Fileformat.Info character search to check out the name and other properties of the character, on the basis of its U+... number.)
The LEFT-TO-RIGHT MARK sets the writing direction of directionally neutral characters. It does not affect e.g. English or Arabic words, but it may mess up text that contains parentheses for example – though for text in English, there should be no confusion in this sense.
But, of course, when text is processed programmatically, as when a web server processes a .htaccess file, they are character data and make a big difference.