<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd"> <HTML> <HEAD> <META HTTP-EQUIV="CONTENT-TYPE" CONTENT="text/html; charset=utf-8"> <TITLE>eSpeak: Adding a Language</TITLE> </HEAD> <BODY LANG="en-GB" DIR="LTR"> <A href="docindex.html">Back</A> <HR> <H2>6. ADDING OR IMPROVING A LANGUAGE</H2> <HR> Most of the work doesn't need any programming knowledge. Just an understanding of the language, an awareness of its features, patience and attention to detail. Wikipedia is a good source of basic phonetic information, eg <a href="http://en.wikipedia.org/wiki/Vowel">http://en.wikipedia.org/wiki/Vowel</a> <P> In many cases it should be fairly easy to add a rough implementation of a new language, hopefully enough to be intelligible.<br> After that it's a gradual process of improvement to: <ul> <li>Make the spelling-to-phoneme translation rules more accurate, including the position of stressed syllables within words. Some languages are easier than others. I expect most are easier than English. <p><li>Improve the sounds of the phonemes. It may be that a phoneme should sound different depending on adjacent sounds, or whether it's at the start or the end of a word, between vowels, etc. This may consist of making small adjustments to vowel and diphthong quality, or adjusting the strength of consonants. Bigger changes may be recording new or replacement consonant sounds, or even writing program code to implement new types of sounds. <p><li>Some common words should be added to the dictionary (the *_list file for the language) with an "unstressed" attribute (eg. in English, words such as "the", "is", "had", "my", "she", "of", "in", "some"), or should be preceded by a short pause (such as "and", "but", "which"), or have other attributes, in order to make the speech flow better. <p><li>Improve the rhythm of the speech by adjusting the relative lengths of vowels in different contexts, eg. stressed/unstressed syllable, or depending on the following phonemes. This is important for making the speech sound good for the language. <p><li>Identify or implement new functions in the program to improve the speech, or to deal with characteristics of the language which are not currently implemented. For example, a different intonation module. </ul> <b><em>If you are interested in working on a language, please contact me so that I can set up the initial data and discuss the features of the language.</em></b> <p> For most of the eSpeak voices, I do not speak or understand the language, and I do not know how it should sound. I can only make improvements as a result of feedback from speakers of that language. If you want to help to improve a language, listen carefully and try to identify individual errors, either in the spelling-to-phoneme translation, the position of stressed syllables within words, or the sound of phonemes, or problems with rhythm and vowel lengths. <HR> <H3>6.1 Language Code</H3> <P>Generally, the language's international <a href="http://en.wikipedia.org/wiki/ISO_639-1">ISO 639-1 code</a> is used to identify the language. It is used in the filenames which contains the language's data. In the examples below the code "<B>en</B>" (English) is used as an example. Replace this with the code of your language.<p> It is possible to have different variants of a language, for example where the sound of some phonemes changed, or where some of the pronunciation rules differ. <HR> <H3>6.2 Phoneme File</H3> <P>You must first decide on the set of phonemes to be used for the language. These should be listed and defined in a phonemes file such as <B>ph_english</B>. A reference to this file is then included at the end of the <B>phonemes,</B> file (the master phoneme file), eg:</P> <PRE> phonemetable en base include ph_english</PRE><P> This example defines a phoneme table "<B>en</B>" which inherits the contents of phoneme table "<B>base</B>". Its contents are found in the file <B>ph_english</B>.</P> <P>The <B>base</B> phoneme table contains definitions of a basic set of consonants, and also some "control" phonemes such as stress marks and pauses. The phoneme table for a language will generally inherit this, or alternatively it may inherit the phoneme table of another language which in turn inherits the <B>base</B> phoneme table.</P> <P>The phonemes file for the language defines those additional phonemes which are not inherited (generally the vowels and diphthongs, plus any additional consonants), or phonemes whose definitions differ from the inherited version (eg. the redefinition of a consonant).</P> <P>Details of the contents of phonemes files are given in <A href="phontab.html">phontab.html</A>.</P> The <B>Compile phoneme data</B> function of the <B>espeakedit</B> program compiles the phonemes files to produce the files <B>espeak-data/phontab</B>, <B>phonindex</B>, and <B>phondata.</B><P> For information on how to analyse recorded sounds of the language and to prepare the corresponding phoneme data, see <a href="editor_if.html">espeakedit</a> and <a href="analyse.html">analysis</a>).<p> For an initial draft a language will often be able to use vowels and consonants which have already been set up for another language. <HR> <H3>6.3 Dictionary Files</H3> <P STYLE="font-weight: medium">Once the language's phonemes have been defined, then pronunciation dictionary data can be produced in order to translate the language's source text into phonemes. This consists of two source files: <B>en_rules</B> (the spelling to phoneme rules) and <B>en_list</B> (an exceptions list, and attributes of certain words). The corresponding compiled data file is <B>espeak-data/en_dict</B> which is produced from <B>en_rules</B> and <B>en_list</B> sources by the command: <B>speak --compile=en</B>.</P> <P STYLE="font-weight: medium">Details of the contents of the dictionary files are given in <A href="dictionary.html">dictionary.html</A>.</P> <P STYLE="font-weight: medium">The <B>en_list</B> file contains not only pronunciation exceptions, but also gives attributes to specific words, Most notable of these are:</P> <P STYLE="font-weight: medium"><B>$u </B>Some common words should be marked as "unstressed" in order to make the speech flow better. These words generally include articles (eg: a, the, this, that), auxillary verbs (eg: is, have, will, can, may), pronouns and possessive adjectives (eg: he, his), some common prepositions (eg: of, to, in, of), some common conjunctions (eg. and, or, if), some common adverbs and adjectives (eg. any, already)</P> <P><B>$pause </B>Some words should be marked to have a short pause before then, in order to produce natural pauses in long sentences. These include conjunctions (eg. and, or, but, however, which) and perhaps some prepositions.</P> <HR> <H3>6.4 Voice File</H3> <P STYLE="font-weight: medium">Each language should have one or more voice files in <B>espeak-data/voices</B>. The filename of the default voice for a language should be the same as the language code.</P> <P STYLE="font-weight: medium">Details of the contants of voice files are given in <A href="voices.html">voices.html</A>.</P> <P STYLE="font-weight: medium">The simplest voice file would contain just a single line to give the language code, eg:</P> <PRE STYLE="margin-bottom: 0.5cm"> language en</PRE><P STYLE="font-weight: medium"> This language code specifies the phoneme table (i.e. <b>phonemetable en</b> and the dictionary (i.e. <B>espeak-data/en_dict</B>) to be used. If needed, these can be overridden by <B>phonemes</B> and <B>dictionary</B> attributes in the voices file.</P> <HR> <H3>6.5 Program Code</H3> <P STYLE="font-weight: medium">The behaviour of the speak program is controlled by various options (eg. whether words are stressed on the first, last, or penultimate syllable). The function <B>SetTranslator()</B> at the start of the <B>tr_languages.cpp</B> file recognizes the language code and sets the appropriate set of options.</P> <P STYLE="font-weight: medium">For a new language, you would add its language code and the required options in <B>SetTranslator()</B>. However, this may not be necessary during testing because most of the options can also be set from the voice file in <B>espeak-data/voices</B>.</P> </BODY> </HTML>
