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 http://en.wikipedia.org/wiki/Vowel.
In many cases it should be fairly easy to add a rough implementation of a new language, hopefully enough to be intelligible. After that it's a gradual process of improvement.
Generally, the language's international ISO 639-1 code is used to identify the language. It is used in the filenames which contain the language's data. In the examples below the code "fr" is used as an example. Replace this with the code of your language.
If the language does not have a 2-letter ISO_639-1 code, then use the 3-letter ISO_639-3 code. Language codes may differ from country codes.
It is possible to have different variants of a language for different dialects. For example the sound of some phonemes are changed, or some of the pronunciation rules differ.
The following files are needed for your language.
The fr_rules and fr_list files are compiled to produce the file espeak-data/fr_dict, which eSpeak uses when it is speaking.
Each language needs a voice file in espeak-data/voices or espeak-data/voices/test. The filename of the default voice for a language should be the same as the language code (eg. "fr" for French).
Details of the contents of voice files are given in voices.html.
The simplest voice file would contain just 2 lines to give the language name and language code, eg:
name french language fr
This language code specifies which phoneme table and dictionary to use (i.e. phonemetable fr and espeak-data/fr_dict) to be used. If needed, these can be overridden by phonemes and dictionary attributes in the voice file. For example you may want to start the implementation of a new language by using the phoneme table of an existing language.
You must first decide on the set of phonemes (vowel and consonant sounds) for the language. These should be defined in a phoneme definition file ph_xxxx, where "ph_xxxx" is the name of your language. A reference to this file is then included at the end of the master phoneme file, phsource/phonemes, eg:
phonemetable fr base include ph_french
This example defines a phoneme table "fr" which inherits the contents of phoneme table "base". Its contents are found in the file ph_french.
The base phoneme table contains definitions of a basic set of consonants, and also some "control" phonemes such as stress marks and pauses. These are defined in phsource/phonemes. The phoneme table for a language will inherit these, or alternatively it may inherit the phoneme table of another language which in turn inherits the base phoneme table.
The phonemes file for the language defines those additional phonemes which are not inherited (generally the vowels and diphthongs, plus any additional consonants that are needed), or phonemes whose definitions differ from the inherited version (eg. the redefinition of a consonant).
Details of phonemes files are given in phontab.html.
The Compile phoneme data function of the espeakedit program compiles the phonemes files of all languages to produce the files espeak-data/phontab, phonindex, and phondata which are used by eSpeak.
For many languages, the consonant phonemes which are already available in eSpeak, together with the available vowel files which can be used to define vowel phonemes, will be sufficient. At least for an initial implementation.
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: fr_rules (the spelling to phoneme rules) and fr_list (an exceptions list, and attributes of certain words). The corresponding compiled data file is espeak-data/fr_dict which is produced from fr_rules and fr_list sources by the command:
espeak --compile=fr.
Or by using the espeakedit program.
Details of the contents of the dictionary files are given in dictionary.html.
The fr_list file contains:
The behaviour of the eSpeak program is controlled by various options such as:
The function SetTranslator() at the start of the source code file tr_languages.cpp recognizes the language code and sets the appropriate options. For a new language, you would add its language code and the required options in SetTranslator(). However, this may not be necessary during testing because most of the options can also be set in the voice file in espeak-data/voices (see Voice files).
Listen carefully to the eSpeak voice. Try to identify what sounds wrong and what needs to be improved.
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.
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.
Recordings of spoken words and phrases can be analysed to try and make eSpeak match a language more closely. Unlike most other (larger and better quality) synthesizers, eSpeak's data is not produced directly from recorded sounds. To use an analogy, it's like a drawing or sketch compared with a photograph. Or vector graphics compared with a bitmap image. It's smaller, less accurate, with less subtlety, but it can sometimes show some aspects of the picture more clearly than a more accurate image.
Spectrum->To_eSpeak) which analysis a voiced sound and produces a file which can be loaded into espeakedit. Details of the modification are in the "praat-mod" directory in the espeakedit package.
The analysis contains a sequence of frames, one per cycle at the speech's fundamental frequency. Each frame is a short time spectrum, together with praat's estimation of the f1 to f5 formant frequencies at the time of that cycle.
I also use Praat's New->Record_mono_sound function to make sound recordings.
Spectrum->To_eSpeak modification (this was named To_Spectrogram2 in earlier versions) to analyse the sound. It produces a file named "spectrum.dat".
Load the "spectrum.dat" file into espeakedit. Espeakedit has two Open functions, File->Open and File->Open2. They are the same, except that they remember different paths. I generally use File->Open2 for reading the "spectrum.dat" file.
The data is displayed in espeakedit as a sequence of spectrum frames (see editor.html).
phsource/vowel and open it into espeakedit. It may be useful to use phsource/vowel/vowelchart as a map to show how vowel files compare with each other. You can select a keyframe from the vowel file and use CTRL-C and CTRL-V to copy the green formant peaks onto a frame of the new spectrum sequence. Then adjust the peaks to match the new frame. Press F1 to hear the sound of the formant peaks in the selected frame.
The F0 peak is provided in order to adjust the correct balance of low frequencies, below the F1 peak. If the sound is too muffled, or conversely, too "thin", try adjusting the amplitude or position of the F0 peak.
Data->Compile_Phoneme_Data from espeakedit's menubar to re-compile the phoneme data.
espeakedit-1.46.02/docs/commands.html 0000664 0000000 0000000 00000027500 11711044451 0017423 0 ustar 00root root 0000000 0000000
Place the "espeak-data" directory in /usr/share as /usr/share/espeak-data.
Alternatively if it is placed in the user's home directory (i.e. /home/<user>/espeak-data)
then that will be used instead.
Some Linux distrubitions (eg. SuSe 10) have version 19 of PortAudio which has a slightly different API. The speak program can be compiled to use version 19 of PortAudio by copying the file portaudio19.h to portaudio.h before compiling.
The speak program may be compiled without using PortAudio, by removing the line
#define USE_PORTAUDIOin the file speech.h.
In this example, "en" is the voice name, "190" is the speed.
Note:
Or just type
espeak
followed by text on subsequent lines. Each line is spoken when
RETURN is pressed.
Use espeak -x to see the corresponding phoneme codes.
1 UTF-8. This is the default.
2 The 8-bit character set which corresponds to the language (eg. Latin-2 for Polish).
4 16 bit Unicode.
Without this option, eSpeak assumes text is UTF-8, but will automatically switch to the 8-bit character set if it finds an illegal UTF-8 sequence.
1 eSpeak uses a click sound to indicate when a word starts with a capital letter, or double click if word is all capitals.
2 eSpeak speaks the word "capital" before a word which begins with a capital letter.
Other values: eSpeak increases the pitch for words which begin with a capital letter. The greater the value, the greater the increase in pitch. Try -k20.
espeak -vafTo use the Afrikaans voice. A modifier after the voice name can be used to vary the tone of the voice, eg:
espeak -vaf+3The variants are
+m1 +m2 +m3 +m4 +m5 +m6 +m7 for male voices and +f1 +f2 +f3 +f4 which simulate female voices by using higher pitches. Other variants include +croak and +whisper.
<voice filename> is a file within the espeak-data/voices directory.
<variant> is a file within the espeak-data/voices/!v directory.
Voice files can specify a language, alternative pronunciations or phoneme sets, different pitches, tonal qualities, and prosody for the voice. See the voices.html file.
Voice names which start with mb- are for use with Mbrola diphone voices, see mbrola.html
Some languages may need additional dictionary data, see languages.html
--punct=".,;?"
<minutes> minutes, at the next sentence boundary.
--voices=mbrola lists the voices which use mbrola diphone voices. These are not included in the default --voices list--voices=variant lists the available voice variants (voice modifiers).
eg: espeak -v en "[[D,Is Iz sVm f@n'EtIk t'Ekst 'InpUt]]"
This command will speak: "This is some phonetic text input".
There are two separate methods for translating words into phonemes:
The utility 'phonemes' are:
| ' | primary stress |
| , | secondary stress |
| % | unstressed syllable |
| = | put the primary stress on the preceding syllable |
| _: | short pause |
| _ | a shorter pause |
| || | indicates a word boundary within a phoneme string |
| | | can be used to separate two adjacent characters, to prevent them from being considered as a multi-character phoneme mnemonic |
The phonemes which are used to represent a language's sounds are based loosely on the Kirshenbaum ascii character representation of the International Phonetic Alphabet www.kirshenbaum.net/IPA/ascii-ipa.pdf
To find the pronunciation of a word, the rules are searched and any which match the letters at the in the word are given a score depending on how many letters are matched. The pronunciation from the best matching rule is chosen. The pointer into the source word is then advanced past those letters which have been matched and the process is repeated until all the letters of the word have been processed.
.L01 b bl br pl pr
.group o
o 0 // "o" is pronounced as [0]
oo u: // but "oo" is pronounced as [u:]
b) oo (k U
"oo" is pronounced as [u:], but when also preceded by "b" and followed by "k", it is pronounced [U].
In the case of a single-letter group, the first character of <match> much be the group letter. In the case of a 2-letter group, the first two characters of <match> must be the group letters. The second and third rules above may be in either .group o or .group oo
Alphabetic characters in the <pre>, <match>, and <post> parts must be lower case, and matching is case-insensitive. Some upper case letters are used in <pre> and <post> with special meanings.
| _^_<language code> | Translate using a different language. |
th (_ _^_EN
indicates that a word which ends in "th" is translated using the English translation rules and spoken with English phonemes.
| _ | Beginning or end of a word (or a hyphen). |
| - | Hyphen. |
| A | Any vowel (the set of vowel characters may be defined for a particular language). |
| C | Any consonant. |
| B H F G Y | These may indicate other sets of characters (defined for a particular language). |
| L<nn> | Any of the sequence of characters defined as a letter group (see 4.3.1 above). |
| D | Any digit. |
| K | Not a vowel (i.e. a consonant or word boundary or non-alphabetic character). |
| X | There is no vowel until the word boundary. |
| Z | A non-alphabetic character. |
| % | Doubled (placed before a character in <pre> and after it in <post>. |
| / | The following character is treated literally. |
Examples of rules:
_) a // "a" at the start of a word
a (CC // "a" followed by two consonants
a (C% // "a" followed by a double consonant (the same letter twice)
a (/% // "a" followed by a percent sign
%C) a // "a" preceded by a double consonants
| @ | Any syllable. |
| & | A syllable which may be stressed (i.e. is not defined as unstressed). |
| V | Matches only if a previous word has indicated that a verb form is expected. |
@@) bi // "bi" preceded by at least two syllables
@@a) bi // "bi" preceded by at least 2 syllables and following 'a'
Note, that matching characters in the <pre> part do not affect the syllable counting.
| @ | A vowel follows somewhere in the word. |
| + | Force an increase in the score in this rule (may be repeated for more effect). |
| S<number> | This number of matching characters are a standard suffix, remove them and retranslate the word. |
| P<number> | This number of matching characters are a standard prefix, remove them and retranslate the word. |
| Lnn | nn is a 2-digit decimal number in the range 01 to 20 Matches with any of the letter sequences which have been defined for letter group nn |
| N | Only use this rule if the word is not a retranslation after removing a suffix. |
| # | (English specific) change the next "e" into a special character "E" |
| $w_alt $w_alt2 $w_alt3 |
Only use this rule if the word is found in the *_list file with the $alt, $alt2 or $alt3 attribute respectively. |
| $p_alt $p_alt2 $p_alt3 |
Only use this rule if the part-word, up to and including the pre and match parts of this rule, is found in the *_list file with the $alt, $alt2 or $alt3 attribute respectively. |
@) ly (_$2 lI // "ly", at end of a word with at least one other
// syllable, is a suffix pronounced [lI]. Remove
// it and retranslate the word.
_) un (@P2 ¬Vn // "un" at the start of a word is an unstressed
// prefix pronounced [Vn]
_) un (i ju: // ... except in words starting "uni"
_) un (inP2 ,Vn // ... but it is for words starting "unin"
S and P must be at the end of the <post> string.
S<number> may be followed by additonal letters (eg. S2ei ). Some of these are probably specific to English, but similar functions could be made for other languages.
| q | query the _list file to find stress position or other attributes for the stem, but don't re-translate the word with the suffix removed. |
| t | determine the stress pattern of the word before adding the suffix |
| d | the previous letter may have been doubled when the suffix was added. |
| e | "e" may have been removed. |
| i | "y" may have been changed to "i." |
| v | the suffix means the verb form of pronunciation should be used. |
| f | the suffix means the next word is likely to be a verb. |
P<number> may be followed by additonal letters (eg. P3v ).
| t | determine the stress pattern of the word before adding the prefix |
| v | the suffix means the verb form of pronunciation should be used. |
Also the list can be used to specify the stress pattern, or other properties, of a word.
If the Pronunciation rules are applied to a word and indicate a standard prefix or suffix, then the word is again looked up in Pronunciation Dictionary List after the prefix or suffix has been removed.
Lines in the dictionary list have the form:
book bUkRather than a full pronunciation, just the stress may be given, to change where it would be otherwise placed by the Pronunciation Rules:
berlin $2 // stress on second syllable
absolutely $3 // stress on third syllable
for $u // an unstressed word
(de jure) deI||dZ'U@rI2 // note || used as a word break in the phoneme stringor to run them together, pronounced as a single word
(of a) @v@or to give them a flag when they occur together
(such as) sVtS||a2z $pause // precede with a pauseHyphenated words in the <language>_list file must also be enclosed within brackets, because the two parts are considered as separate words.
| _^_<language code> | Translate using a different language. See explanation in 4.3.3 above. |
| $u | The word is unstressed. In the case of a multi-syllable word, a slight stress is applied according to the default stress rules. |
| $u1 | The word is unstressed, with a slight stress on its 1st syllable. |
| $u2 | The word is unstressed, with a slight stress on its 2nd syllable. |
| $u3 | The word is unstressed, with a slight stress on its 3rd syllable. |
| $u+ $u1+ $u2+ $u3+ | As above, but the word has full stress if it's at the end of a clause. |
| $1 | Primary stress on the 1st syllable. |
| $2 | Primary stress on the 2nd syllable. |
| $3 | Primary stress on the 3rd syllable. |
| $4 | Primary stress on the 4th syllable. |
| $5 | Primary stress on the 5th syllable. |
| $6 | Primary stress on the 6th syllable. |
| $7 | Primary stress on the 7th syllable. |
| $pause | Ensure a short pause before this word (eg. for conjunctions such as "and", some prepositions, etc). |
| $brk | Ensure a very short pause before this word, shorter than $pause (eg. for some prepositions, etc). |
| $only | The rule does not apply if a prefix or suffix has already been removed. |
| $onlys | As $only, except that a standard plural ending is allowed. |
| $stem | The rule only applies if a suffix has already been removed. |
| $strend | Word is fully stressed if it's at the end of a clause. |
| $strend2 | As $strend, but the word is also stressed if followed only by unstressed word(s). |
| $unstressend | Word is unstressed if it's at the end of a clause. |
| $atend | Use this pronunciation if it's at the end of a clause. |
| $double | Cause a doubling of the initial consonant of the following word (used for Italian). |
| $capital | Use this pronunciation if the word has initial capital letter (eg. polish v Polish). |
| $allcaps | Use this pronunciation if the word is all capitals. |
| $dot | Ignore a . after this word even when followed by a capital letter (eg. Mr. Dr. ). |
| $hasdot | Use this pronunciation if the word is followed by a dot. (This attribute also implies $dot). |
| $sentence | The rule only applies if the clause includes end-of-sentence (i.e. it is not terminated by a comma). For example, "$atend $sentence" means that the rule only applies at the end of a sentence. |
| $abbrev | This has two meanings. 1. If there is no phoneme string: Speak the word as individual letters, even if it contains a vowel (eg. "abc" should be spoken as "a" "b" "c"). 2. If there is a phoneme string: This word is capitalized because it is an abbreviation and capitalization does not indicate emphasis (if the "emphasize all-caps" is on). |
| $accent | Used for the pronunciation of a single alphabetic character. The character name is spoken as the base-letter name plus the accent (diacritic) name. eg. It can be used to specify that "â" is spoken as "a" "circumflex". |
| $combine | This word is treated as though it is combined with the following word with a hyphen. This may be subject to fuither conditions for certain languages. |
| $alt $alt2 $alt3 | These are language specific. Their use should be described in the language's **_list file |
| $verb | Use this pronunciation if it's a verb. |
| $noun | Use this pronunciation if it's a noun. |
| $past | Use this pronunciation if it's past tense. |
| $verbf | The following word is probably is a verb. |
| $verbsf | The following word is probably is a if it has an "s" suffix. |
| $nounf | The following word is probably not a verb. |
| $pastf | The following word is probably past tense. |
| $verbextend | Extend the influence of $verbf and $verbsf. |
The dictionary list is searched from bottom to top. The first match that satisfies any conditions is used (i.e. the one lowest down the list). So if we have:
to t@ // unstressed version
to tu: $atend // stressed version
then if "to" is at the end of the clause, we get [tu:], if not then we get [t@].
Use the attribute $text eg.
cough coff $text
Alternatively, use the command $textmode on a line by itself to turn this on for all subsequent entries in the file, until it's turned off by $phonememode. eg.
$textmode
cough coff
through threw
$phonememode
This feature cannot be used for the special entries in the _list files which start with an underscore, such as numbers.Currently "textmode" entries are only recognized for complete words, and not for for stems from which a prefix or suffix has been removed (eg. the word "coughs" would not match the example above).
Conditional rules have ? and a condition number at the start if the line in the _rules or _list file. This means that the rule only applies of that condition number is specified in a dictrules line in the voice file.
If the rule starts with ?! then the rule only applies if the condition number is not specified in the voice file. eg.
?3 can't kant // only use this if the voice has: dictrules 3 ?!3 rather rA:D3 // only use if the voice doesn't have: dictrules 3
_a eI
langopts.numbers option. This constructs spoken numbers from fragments according to various options which can be set for each language. The number fragments are given in the _list file.
| _0 to _9 | The numbers 0 to 9 |
| _13 | etc. Any pronunciations which are needed for specific numbers in the range _10 to _99 |
| _2X _3X | Twenty, thirty, etc., used to make numbers 10 to 99 |
| _0C | The word for "hundred" |
| _1C _2C | Special pronunciation for one hundred, two hundred, etc., if needed. |
| _1C0 | Special pronunciation (if needed) for 100 exactly |
| _0M1 | The word for "thousand" |
| _0M2 | The word for "million" |
| _0M3 | The word for 1000000000 |
| _1M1 _2M1 | Special pronunciation for one thousand, two thousand, etc, if needed |
| _0and | Word for "and" when speaking numbers (eg. "two hundred and twenty"). |
| _dpt | Word spoken for the decimnal point/comma |
| _dpt2 | Word spoken (if any) at the end of all the digits after a decimal point. |
.replace
ô ő // (Hungarian) allow the use of o-circumflex instead of o-double-accute
û ű
cx ĉ // (Esperanto) allow "cx" as an alternative to c-circumflex
fi fi // replace a single character ligature by two characters
espeakedit-1.46.02/docs/docindex.html 0000664 0000000 0000000 00000004702 11711044451 0017416 0 ustar 00root root 0000000 0000000
|
|
eSpeak - Documents |
| Home |
Voice FilesVoice files specify a language and other characteristics of a voice.Mbrola VoiceseSpeak can be used as a front-end for Mbrola diphone voices.Pronunciation Dictionary
Adding a LanguageHow to add or improve a language.PhonemesThe list of phoneme mnemonics for English, for use in the Pronunciation Dictionary.Phoneme TablesThe tables of the phonemes used by each language, with their properties and sound production.IntonationDifferent intonation "tunes" may be defined for different languages for clauses which end in full-stop, comma, question-mark, and exclamation-mark.eSpeak Libary APIAPI definition and header file for a shared library version of eSpeak.Markup tagsSSML (Speech Synthesis Markup Language) and HTML tags recognized by eSpeak.The espeakedit programGUI software to edit vowel files and to compile the phoneme data for use by eSpeak. |
It has two main functions:
For more detailed information on analysing sound recordings and preparing phoneme definitions and keyframe data see analyse.html (to be written).
NOTE: espeakedit can be used from the command line to compile the phoneme data, with the command: espeakedit --compile
If changes are made in the Prosody tab, then clicking Speak will speak the modified prosody while Translate will revert to the default prosody settings for the text.
To enter phonetic symbols (Kirschenbaum encoding) in the top left text window, enclose them within [[ ]].
The selected frame(s) are shown with a red border. The selected formant peak is also indicated by an equals ("=") sign next to its number in the "Spect" panel to the right of the window.
The selected formant peak is shown with a red triangle under the peak.
Keyframes are shown with a pale yellow background. A keyframe is any frame with any formant peaks which are not zero height. If all formant peaks become zero height, the frame is no longer a keyframe. If you increase a peak's height the frame becomes a keyframe.
|
|
eSpeak text to speech |
| Usage |
eSpeak is a compact open source software speech synthesizer for English and other languages, for Linux and Windows.
http://espeak.sourceforge.net
eSpeak uses a "formant synthesis" method. This allows many languages to be provided in a small size. The speech is clear, and can be used at high speeds, but is not as natural or smooth as larger synthesizers which are based on human speech recordings. eSpeak is available as:
I regularly use eSpeak to listen to blogs and news sites. I prefer the sound through a domestic stereo system rather than small computer speakers, which can sound rather harsh. Languages. The eSpeak speech synthesizer supports several languages, however in many cases these are initial drafts and need more work to improve them. Assistance from native speakers is welcome for these, or other new languages. Please contact me if you want to help. eSpeak does text to speech synthesis for the following languages, some better than others. Afrikaans, Albanian, Armenian, Cantonese, Catalan, Croatian, Czech, Danish, Dutch, English, Esperanto, Finnish, French, German, Greek, Hindi, Hungarian, Icelandic, Indonesian, Italian, Kurdish, Latvian, Lojban, Macedonian, Mandarin, Norwegian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Spanish, Swahili, Swedish, Tamil, Turkish, Vietnamese, Welsh. The latest development version is at: espeak.sf.net/test/latest.html. espeakedit is a GUI program used to prepare and compile phoneme data. It is now available for download. Documentation is currently sparse, but if you want to use it to add or improve language support, let me know. History. Originally known as speak and originally written for Acorn/RISC_OS computers starting in 1995. This version is an enhancement and re-write, including a relaxation of the original memory and processing power constraints, and with support for additional languages. |
Named tunes are defined in the text file: phsource/intonation. This file must be compiled for use by eSpeak by using the espeakedit program, using the menu option: Compile -> Compile intonation data.
tunes statement in a voice file in espeak-data/voices. eg:
tunes s1 c1 q1 e1
It's parameters are four tune names which are used for clauses which end in:
A clause consists of the following parts:
phsource/intonation.
tune s1 prehead 46 57 headenv fall 16 head 4 80 55 -8 -5 headextend 0 63 38 13 0 nucleus fall 70 18 24 12 nucleus0 fall 64 8 endtune
It contains:
tune name can be used in a tunes statements in voice files.
onset and headlast syllables if these are specified. height gives a pitch range for the envelope.
start pitch and end pitch give a pitch path for the stressed syllables of the head. steps is the maximum number of stressed syllables for which this applies. If there are additional stressed syllables, then the headextend statement is used for them.
unstressed start and unstressed end give a pitch path for unstressed syllables between two stressed syllables. Their values are relative to the pitch of the previous stressed syllable. Values are usually negative, meaning that the unstressed syllables have lower pitch than the previous stressed syllable.
steps, then percentage list is used. It contains up to 8 numbers which are used repeatedly for the additional stressed syllables. A value of 0 corresponds to the lower the start pitch and end pitch values of the head statement. 100 corresponds to the higher value. Negative values and values greater than 100 are allowed.
tail start and tail end give a pitch path for the unstressed syllables which are after the last stressed syllable.
nucleus if there are no unstressed syllables after the last stressed syllable. In this case, the pitch changes of the nucleus and the tail and both included in the nucleus.
The following attributes may also be included:
onset statement is present, then the head statement used for the stressed syllables after the first.
The sound of some phonemes may be poorly implemented, particularly [r] since I'm English and therefore unable to make a "proper" [r] sound.
A major factor is the rhythm or cadance. An Italian speaker told me the Italian voice improved from "difficult to understand" to "good" by changing the relative length of stressed syllables. Identifying unstressed function words in the xx_list file is also important to make the speech flow well. See Adding or Improving a Language
charset 5will mean that ISO 8859-5 is used as the 8-bit character set rather than KOI8-R.
In the case of a language which uses a non-Latin character set (eg. Greek or Russian) if the text contains a word with Latin characters then that particular word will be pronounced using English pronunciation rules and English phonemes. Speaking entirely English text using a Greek or Russian voice will sound OK, but each word is spoken separately so it won't flow properly.
Sample texts in various languages can be found at http://<language>.wikipedia.org and www.gutenberg.org
espeak-data/voices directory.
You can select one of these with the -v <voice filename> parameter to the
speak command, eg:
espeak -vafto speak using the Afrikaans voice.
Language voices generally start with the 2 letter ISO 639-1 code for the language. If the language does not have an ISO 639-1 code, then the 3 letter ISO 639-3 code can be used.
For details of the voice files see Voices.
Alternatively there are some preset voice variants which can be applied to any of the language voices, by appending + and a variant name. Their effects are defined by files in espeak-data/voices/!v.
The variants are +m1 +m2 +m3 +m4 +m5 +m6 +m7 for male voices, +f1 +f2 +f3 +f4 +f5 for female voices, and +croak +whisper for other effects. For example:
espeak -ven+m3The available voice variants can be listed with:
espeak --voices=variant
espeakedit-1.46.02/docs/mbrola.html 0000664 0000000 0000000 00000016522 11711044451 0017100 0 ustar 00root root 0000000 0000000
eSpeak can be used as a front-end to Mbrola. It provides the spelling-to-phoneme translation and intonation, which Mbrola then uses to generate speech sound.
The eSpeak voices which use Mbrola are named as:
mb-xxx
where xxx is the name of a Mbrola voice (eg. mb-en1 for the Mbrola "en1" English voice). These voice files are in eSpeak's directory espeak-data/voices/mbrola.
The installation instructions below use the Mbrola voice "en1" as an example. You can use other mbrola voices for which there is an equivalent eSpeak voice in espeak-data/voices/mbrola.
There are some additional eSpeak Mbrola voices which speak English text using a Mbrola voice for a different language. These contain the name of the Mbrola voice with a suffix -en. For example, the voice mb-de4-en will speak English text with a German accent by using the Mbrola de4 voice.
C:/Program Files/eSpeak/espeak-data/mbrola.
mbrola-linux-i386 to
mbrola somewhere in your executable path (eg. /usr/bin/mbrola ).
/usr/share/mbrola/en1.eSpeak will look for mbrola voices firstly in espeak-data/mbrola and then in /usr/share/mbrola
espeak -v mb-en1 "Hello world"
To generate mbrola phoneme data (.pho file) you can use:
espeak -v mb-en1 -q --pho "Hello world"
or
espeak -v mb-en1 -q --pho --phonout=out.pho "Hello world"
espeak-data/voices/mbrola. They contain a line:mbrola <voice> <translation>
mbrola en1 en1_phtrans
espeak-data/mbrola_ph
phsource/mbrola, see below.
The source files are in phsource/mbrola. These are compiled using the espeakedit program (Compile->Compile mbrola phonemes list) to produce data files in espeak-data/mbrola_ph which are used by eSpeak.
Each line in the mbrola phoneme translation file contains:
<control> <espeak ph1> <espeak ph2> <percent> <mbrola ph1> [<mbrola ph2>]
NULL, then the match only occurs if this field matches the next phoneme. If control bit 1 is set, then the previous rather than the next phoneme is matched. This field may also have the following values:VWL matches any Vowel phoneme.
NULL.
The file dictsource/dict_phonemes lists the eSpeak phonemes which are used for each language. Translations for all these should be given in the mbrola phoneme translation file. In addition, some phonemes which are referenced from phoneme files (eg. phsource/ph_language, phsource/phonemes) in lines such as:
beforenotvowel l/ reduceto a# 0should also be included, even though they don't appear in
dictsource/dict_phonemes.
If the language's *_list or *_rules files includes rules to speak words "as English" the mbrola phoneme translation file should include rules which translate English phonemes into near equivalents, so that they can spoken by the mbrola voice. espeakedit-1.46.02/docs/phonemes.html 0000664 0000000 0000000 00000012232 11711044451 0017434 0 ustar 00root root 0000000 0000000
In most cases different languages inherit the same basic set of consonants. They can add to these or modify them as needed.
The phoneme mnemonics are based on the scheme by Kirshenbaum which represents International Phonetic Alphabet symbols using ascii characters. See: www.kirshenbaum.net/IPA/ascii-ipa.pdf.
Phoneme mnemonics can be used directly in the text input to espeak. They are enclosed within double square brackets. Spaces are used to separate words, and all stressed syllables must be marked explicitly. eg:
[[D,Is Iz sVm f@n'EtIk t'Ekst 'InpUt]]
[p] | [b] | ||
[t] | [d] | ||
[tS] | church | [dZ] | judge |
[k] | [g] | ||
| |||
[f] | [v] | ||
[T] | thin | [D] | this |
[s] | [z] | ||
[S] | shop | [Z] | pleasure |
[h] | |||
| |||
[m] | [n] | ||
[N] | sing | ||
[l] | [r] | red (Omitted if not immediately followed by a vowel). | |
[j] | yes | [w] | |
| |||
| Some Additional Consonants | |||
[C] | German ich | [x] | German buch |
[l^] | Italian gli | [n^] | Spanish ñ |
In rhotic accents, such as General American, the phonemes [3:], [A@], [e@], [i@], [O@], [U@] include the "r" sound.
[@]
| alpha | schwa |
[3]
| better | rhotic schwa. In British English this is the same as [@], but it includes 'r' colouring in American and other rhotic accents. In these cases a separate [r] should not be included unless it is followed immediately by another vowel.
|
[3:] | nurse | |
[@L] | simple | |
[@2] | the | Used only for "the". |
[@5] | to | Used only for "to". |
| ||
[a] | trap | |
[aa] | bath | This is [a] in some accents, [A:] in others.
|
[a2] | about | This may be [@] or may be a more open schwa.
|
[A:] | palm | |
[A@] | start | |
| ||
[E] | dress | |
[e@] | square | |
| ||
[I] | kit | |
[I2] | intend | As [I], but also indicates an unstressed syllable.
|
[i] | happy | An unstressed "i" sound at the end of a word. |
[i:] | fleece | |
[i@] | near | |
| ||
[0] | lot | |
| ||
[V] | strut | |
| ||
[u:] | goose | |
[U] | foot | |
[U@] | cure | |
| ||
[O:] | thought | |
[O@] | north | |
[o@] | force | |
| ||
[aI] | price | |
[eI] | face | |
[OI] | choice | |
[aU] | mouth | |
[oU] | goat | |
[aI@] | science | |
[aU@] | hour |
[e] | German eh, French é |
[o] | German oo, French o |
[y] | German ü, French u |
[Y] | German ö, French oe |
Generally each language has its own phoneme table, although additional phoneme tables can be used for different voices within the language. These alternatives are referenced from Voice files.
A phoneme table does not need to define all the phonemes used by a language. It can inherit the phonemes from a previously defined phoneme table. For example, a phoneme table may redefine (or add) some of the vowels that it uses, but inherit most of its consonants from a standard set.
The source files for the phoneme data are in the "phsource" directory in the espeakedit download package. "Vowel files", which are referenced in FMT(), VowelStart(), and VowelEnding() instructions are made using the espeakedit program.
In addition to phoneme definitions, the phoneme file can contain the following:
A phoneme table contains a list of phoneme definitions. Each starts with the keyword phoneme and the phoneme name (this is the name used in the pronunciation rules in a language's *_rules and *_list files), and ends with the keyword endphoneme. For example:
phoneme aI
vowel
starttype #a endtype #i
length 230
FMT(vowels/ai)
endphoneme
phoneme s
vls alv frc sibilant
voicingswitch z
lengthmod 3
Vowelin f1=0 f2=1700 -300 300 f3=-100 80
Vowelout f1=0 f2=1700 -300 250 f3=-100 80 rms=20
IF nextPh(isPause) THEN
WAV(ufric/s_)
ELIF nextPh(p) OR nextPh(t) OR nextPh(k) THEN
WAV(ufric/s!)
ENDIF
WAV(ufric/s)
endphoneme
A phoneme definition contains both static properties and executed instructions. The instructions may contain conditional statements, so that the effect of the phoneme may be different depending on adjacent phonemes, whether the syllable is stressed, etc.
The instructions of a phoneme are interpreted in two different phases. In the first phase, the instructions may change the phoneme and replace it by a different phoneme. In the second phase, instructions are used to produce the sound for the phoneme.
The import_phoneme statement can be used to copy a previously defined phoneme from a specified phoneme table. For example:
phoneme t
import_phoneme base/t[
endphoneme
means: phoneme t in this phoneme table is a copy of phoneme t[ from phoneme table "base". A length instruction can be used after import_phoneme to vary the length from the original.
| vowel | |
| liquid | semi-vowels, such as: r, l, j, w |
| nasal | nasal eg: m, n, N |
| stop | stop eg: p, b, t, d, k, g |
| frc | fricative eg: f, v, T, D, s, z, S, Z, C, x |
| afr | affricate eg: tS, dZ |
| pause | |
| stress | used for stress symbols, eg: ' , = % |
| virtual | Used to represent a class of phonemes. |
| vls | (C) voiceless eg. p, t, k, f, s |
| vcd | (C) voiced eg. b, d, g, v, z |
| sibilant | (C) eg: s, z, S, Z, tS, dZ |
| palatal | (C) A palatal or palatalized consonant. |
| rhotic | (C) An "r" type consonant. |
| unstressed | (V) This vowel is always unstressed, unless explicitly marked otherwise. |
| nolink | Prevent any linking from the previous phoneme. |
| nopause | Used in a liquid or nasal phoneme to prevent eSpeak inserting a short pause if a word starts with this phoneme and the previous word ends with a vowel. |
| trill | (C) Apply trill to the voicing. |
| blb | bi-labial | ldb | labio-dental | dnt | dental |
| alv | alveolar | rfx | retroflex | pla | palato-alveolar |
| pal | palatal | vel | velar | lbv | labio-velar |
| uvl | uvular | phr | pharyngeal | glt | glottal |
length_mods table in the CalcLengths() function in the eSpeak program.
During the first phase of phoneme interpretation, an instruction which causes a change to a different phoneme will terminate the instructions. During the second phase, FMT() and WAV() instructions will terminate the instructions.
phsource/ which will be played to produce the sound. This method is used for unvoiced consonants. <wavefile> does not include a .WAV filename extension, although the file to which it refers may or may not have one.WAV(ufric/s, 50) means: play file 'ufric/s.wav' at 50% amplitude.
phsource/) which defines how to generate the sound (a vowel or voiced consonant) from a sequence of formant values. Vowel files are made using the espeakedit program.
IF <condition> THEN
<statements>
ENDIF
or more generally:
IF <condition> THEN
<statements>
ELIF <condition> THEN
<statements>
...
ELSE
<statements>
ENDIF
where the ELSE and multiple ELSE parts are optional.
Multiple conditions may be joined with AND or OR, but not a mixture of ANDs and ORs.
Condition Can be:
nextPh(isPause2) is used to test whether the next phoneme is not a vowel or liquid consonant within the same word.
[p] [t] [s].
[l] [j] [n].
[b] [g] [v] [z].