voices.md 14 KB

Voice and Language Files

A Voice file specifies a language (and possibly a language variant or dialect) together with various attributes that affect the characteristics of the voice quality and how the language is spoken.

Voice files are located in the espeak-ng-data/voices directory, and are grouped by the ISO 639-5 language family of the language being specified in the voice files. See also Wikipedia's List of language families for more details.

The default voice is used if none is specified in the speak command. You can copy your preferred voice to "default" so you can use the speak command without the need to specify a voice.

Identification Attributes

name

name <name>

A name given to this voice.

language

language <language code> [<priority>]

NOTE: This attribute is mandatory and should appear before the other attributes which are listed below.

It selects the default behaviour and characteristics for the language, and sets default values for "phonemes", "dictionary" and other attributes.

The <language code> is a valid BCP 47 language tag. The list of valid tags originate from various standards and have been combined into the IANA Language Subtag Registry. For example:

  • de (German) -- The ISO 639-1 2-letter language code for the language.

NOTE: BCP 47 uses ISO 639-1 codes for languages that are allocated 2-letter codes (e.g. using en instead of eng).

  • yue (Cantonese) -- The ISO 639-3 3-letter language codes for the language.

  • ta-Arab (Tamil written in the Arabic alphabet) -- The ISO 15924 4-letter script code.

NOTE: Where the script is the primary script for the language, the script tag should be omitted.

  • es-419 (Spanish (Latin America)) -- The UN M.49 3-number region codes.

  • fr-CA (French (Canada)) -- Using the ISO 3166-2 2-letter region codes.

  • en-GB-scotland (English (Scotland)) -- This is using the BCP 47 variant tags.

  • en-GB-x-rp (English (Received Pronunciation)) -- This is using the bcp47-extensions language tags for accents that cannot be described using the available BCP 47 language tags.

NOTE: If the accent you are trying to describe cannot be specified using the above system, raise an issue in the bcp47-data project and a private use tag will be defined for that accent.

The optional <priority> value gives the preference of this voice compared with others for the specified language. A low value indicates a more preferred voice. The default value is 5.

More than one language line may be present. A voice may be selected for other related languages (variants which have the same initial 2 letter language code as the specified language), but it will be less preferred for these. Different language variants may be specified by additional language lines in order to indicate that this is a preferred voice for them also. E.g.

language en-GB-x-gbclan
language en

indicates that this is voice is for the en-GB-x-gbclan dialect, but it is also a main choice when a general en language is specified. Without the second language line, it would be disfavoured from en for being a more specialised voice.

gender

gender <gender> [<age>]

This attribute is only a label for use in voice selection. It doesn't change the sound of the voice.

  • <gender> may be male, female, or unknown.
  • <age> is optional and gives an age in years.

Maintenance Attributes

maintainer

maintainer <maintaner>

Specifies the person responsible for updating the voice and associated language files. If no maintainer is specified, the voice does not currently have an active maintainer.

status

status <status level>

Specifies how advanced the support for the language is:

  • testing -- Provides the basic level of support, with basic pronunciation rules. Requires feedback from a native speaker to help improve the voice.

  • mature -- The language is well supported. It has been improved by, or with support from, a native speaker.

Voice Attributes

pitch

pitch <base> <range>

Two integer values. The first gives a base pitch to the voice (value in Hz) The second controls the range of pitches used by the voice. Setting it equal to the base pitch will give a monotone. The default values are 82 118.

formant

formant <number> <frequency> <strength> <width> <freq_add>

Systematically adjusts the frequency, strength, and width of the resonance peaks of the voice. Values are percentages of the default values. Changing these affects the tone/quality of the voice.

freq_add

Adds a constant value (in Hz) to the frequency of the formant peak. The value may be negative.

  • Formants 1,2,3 are the standard three formants which define vowels.
  • Formant 0 is used to give a low frequency component to the sounds, of frequency lower than F1.
  • Formants 4,5 are higher than F3. They affect the quality of the voice.
  • Formants 6,7,8 are weak, high frequency, additions to vowels to give a clearer sound.

echo

echo <delay> <amplitude>

Parameter 1 gives the delay in mS (0 to 250mS).

Parameter 2 gives the echo amplitude (0 to 100).

Adding some echo can give a clearer or more interesting sound, especially when listening through a domestic stereo sound system, rather than small computer speakers.

tone

Controls the tone of the sound.

tone is followed by up to 4 pairs of <frequency> <amplitude> which define a frequency response graph. Frequency is in Hz and amplitude is in the range 0 to 255 The default is:

tone 600 170  1200 135  2000 110

This means that from frequency 0Hz to 600Hz the amplitude is 170 From 600Hz to 1200Hz the amplitude decreases from 170 to 135, then decreases to 110 at 2000Hz and remains at 110 at higher frequencies. This adjustment applies only to voiced sounds such as vowels and sonorant consonants (such as [n] and [l]). Unvoiced sounds such as [s] are unaffected.

This tone statement can also appear in espeak-ng-data/config, in which case it applies to all voices which don't have their own tone statement.

flutter

flutter <value>

Default value: 100.

Adds pitch fluctuations to give a wavering or older-sounding voice. A large value (eg. 20) makes the voice sound "croaky".

roughness

roughness <value>

Default value: Range 0 - 7

Reduces the amplitude of alternate waveform cycles in order to make the voice sound creaky.

voicing

voicing <value>

Default value: 100

Adjusts the strength of formant-synthesized sounds (vowels and sonorant consonants).

consonants

consonants <value> <value>

Default values: 100, 100

Adjusts the strength of noise sounds which are used in consonants. The first value is the strength of unvoiced consonants such as "s" and "t".

The second value is the strength of the noise component of voiced consonants such as "z" and "d".

breath

breath <up to 8 integer values>

Default values: 0.

Adds noise which corresponds to the formant frequency peaks. The values give the strength of noise for each formant peak (formants 1 to 8).

Use together with a low or zero value of the voicing attribute to make a "wisper". For example:

breath   75 75 60 40 15 10
breathw  150 150 200 200 400 400
voicing  18
flutter  20
formant   0 100 0 100   // remove formant 0 

breathw

breathw <up to 8 integer values>

These values give bandwidths of the noise peaks of the breath attribute. If breathw values are not given, then suitable default values will be used.

speed

speed <value>

Default value 100

Adjusts the speaking speed by a percentage of the default rate. This can be used if a language voice seems faster or slower compared to other voices.

words

words <integer value> <integer value>

The first parameter puts a pause between all words, the value can be between 0 and 4.

The second parameter adds a short pause if a word ends with a vowel and the next words starts with a vowel. Values may be:

  • 0: no pause
  • 1: no pause, but the two vowels are kept separate
  • 2: short pause

For example:

words 0 1

will put a short pause between two words where the first word end with a vowel and the next start with a vowel.

Language Attributes

phonemes

phonemes <name>

Specifies which set of phonemes to use from those contained in the phontab, phonindex, and phondata data files. This is a phonemetable name as given in the "phoneme" source file.

This parameter is usually not needed as it is set by default to the first two letters of the "language" parameter. However, different voices of the same language can use different phoneme sets, to give different accents.

dictionary

dictionary <name>

Specifies which pair of dictionary files to use. For example, en indicates that speak-data/en_dict should be used to translate from words to phonemes. This parameter is usually not needed as it is set by default to the value of the "language" parameter.

dictrules

dictrules <list of rule numbers>

Gives a list of conditional dictionary rules which are applied for this voice. Rule numbers are in the range 0 to 31 and are specific to a language dictionary. They apply to rules in the language's *_rules dictionary file and also its *_list exceptions list. See Text to Phoneme Translation.

replace

replace <flags> <phoneme> <replacement phoneme>

Replace a phoneme by another whenever it occurs.
<replacement phoneme> may be NULL.

Flags: bit 0: replacement only occurs on the final phoneme of a word.
Flags: bit 1: replacement doesn't occur in stressed syllables.

e.g.

replace  0  h  NULL      // drops h's
replace  0  V  U         // replaces vowel in 'strut' by that in 'foot'
                         // as occurs in northern British English
replace  3  N  n         // change 'fishing' to 'fishin' etc.
                         // (only the last phoneme of a word, only in unstressed syllables)

The phoneme mnemonics can be defined for each language, but some are listed in Phonemes.

stressRule

stressRule <4 integer values>

Four integer parameters. These correspond to:

  1. langopts->stress_rule (values in translate.h)
  2. langopts->stress_flags
  3. langopts->unstressed_wd1 (stress for $u word of 1 syllable)
  4. langopts->unstressed_wd2 (stress for $u word of >1 syllable)

If a value is not given, it defaults to zero. For example: "stressRule 2" is equal to "stressRule 2 0 0 0"

stressLength

stressLength <8 integer values>

Eight integer parameters. These control the relative lengths of the vowels in stressed and unstressed syllables.

  • 0 unstressed. Unstressed syllable in word (syllable with % mark or without any stress mark).
  • 1 diminished. Its use depends on the language. In English it's used for unstressed syllables (marked with %%) within multisyllabic words. In other languages it is final unstressed syllable.
  • 2 secondary stress (syllable with marked with ,).
  • 3 words marked as "unstressed" in the dictionary.
  • 4 not currently used
  • 5 not currently used
  • 6 stressed syllable (the main syllable in stressed words marked with ').
  • 7 tonic syllable (by default, the last stressed syllable in the clause).

stressAdd

stressAdd <8 integer values>

Eight integer parameters. These are added to the voice's corresponding stressLength values. They are used in the voice variant files in espeak-ng-data/voices/!v to give some variety. Negative values may be used.

stressAmp

stressAmp <8 integer values>

Eight integer parameters. These control the relative amplitudes of the vowels in stressed and unstressed syllables (see stressLength above). The general default values are: 16, 16, 20, 20, 20, 24, 24, 22, although these defaults may be different for particular languages.

intonation

intonation <param1>
  • 1 -- Default.
  • 2 -- Less intonation.
  • 3 -- Less intonation, and comma does not raise the pitch.
  • 4 -- Pitch rises (rather than falls) at the end of sentence.

dictmin

dictmin <value>

Used for some languages to detect if additional language data is installed. If the size of the compiled dictionary data for the language (the file espeak-ng-data/*_dict) is less than this size then a warning is given.

letterVowel

letterVowel <letter>

Used for some languages to handle a certain letter as a vowel instead of consonant.