eSpeak NG is an open source speech synthesizer that supports more than hundred languages and accents.

Phonemes

IPA Phonemes
Features
Properties
- Tones
References

Evan Kirshenbaum created an ASCII transcription of the International Phonetic Alphabet (IPA)^{[1], [2]}. As well as using ASCII characters for specific IPA phonemes, this transcription provides a set of 3-letter feature abbreviations allowing a phoneme to be described as a sequence of features.

Kirshenbaum uses features that align with the way the IPA phoneme charts are organised. This document uses the Kirshenbaum features as a base to work from, but extends them to describe the underlying phonetic processes in a way that is consistent between languages and voices.

This document is grouped into three sections. The first section displays the IPA charts using the feature names instead of their names, showing the IPA phoneme at that position in the chart. This makes it easier to look up the features for a given IPA phoneme.

The second section lists the features and their associated name. The Wikipedia IPA^[4] article can be used as a starting point into the various phonetic topics contained in this document.

The third section describes the properties (named values) used to describe the phonemes. These, in addition to the features, should allow all possible phonemes from any language to be described.

The goal of this document is not to provide a detailed guide on phonetics. Instead, it is designed to be a transcription guide on how to specify phonemes in a language or voice so that the narrow transcriptions are consistent between the two.

IPA Phonemes

Consonants

	`blb`		`lbd`		`dnt`		`alv`		`pla`		`rfx`		`alp`		`pal`		`vel`		`uvl`		`phr`		`glt`
`nas`		m		ɱ				n				ɳ				ɲ		ŋ		ɴ
`stp`	p	b					t	d			ʈ	ɖ			c	ɟ	k	ɡ	q	ɢ	ʡ		ʔ
`sib` `afr`							t͡s	d͡z	t͡ʃ	d͡ʒ	ʈ͡ʂ	ɖ͡ʐ	t͡ɕ	d͡ʑ
`afr`	p͡ɸ	b͡β	p̪͡f	b̪͡v	t͡θ	d͡ð									c͡ç	ɟ͡ʝ	k͡x	ɡ͡ɣ	q͡χ	ɢ͡ʁ	ʡ͡ħ	ʡ͡ʕ	ʔ͡h
`lat` `afr`							t͡ɬ	d͡ɮ
`sib` `frc`							s	z	ʃ	ʒ	ʂ	ʐ	ɕ	ʑ
`frc`	ɸ	β	f	v	θ	ð									ç	ʝ	x	ɣ	χ	ʁ	ħ	ʕ	h	ɦ
`lat` `frc`							ɬ	ɮ
`apr`				ʋ				ɹ				ɻ				j		ɰ
`lat` `apr`								l				ɭ				ʎ		ʟ
`flp`				ⱱ				ɾ				ɽ
`lat` `flp`								ɺ
`trl`		ʙ						r												ʀ	ʜ	ʢ
`clk`	ʘ				ǀ		ǃ						ǂ
`lat` `clk`							ǁ
`imp`		ɓ						ɗ								ʄ		ɠ		ʛ
`ejc`	pʼ						tʼ				ʈʼ				cʼ		kʼ		qʼ		ʡʼ
`ejc` `frc`	fʼ				θʼ		sʼ		ʃʼ		ʂʼ						xʼ		χʼ
`lat` `ejc` `frc`							ɬʼ

Symbols to the left have a vls phonation, and to the right have mdv phonation.

Other Symbols

Symbol	Alternative	Features
ʍ	ɰ̊ʷ	`vls` `vel` `ptr` `apr`
w	ɰʷ	`mdv` `vel` `ptr` `apr`
ɥ	jʷ	`mdv` `pal` `ptr` `apr`
ɧ		`vls` `vzd` `pla` `frc`
ɫ		`mdv` `fzd` `alv` `lat` `apr`
ɚ		`unr` `mid` `cnt` `rzd` `vwl`
ɝ		`unr` `lmd` `cnt` `rzd` `vwl`
k͡p		`vls` `lbv` `stp`
ɡ͡b		`mdv` `lbv` `stp`
ŋ͡m		`mdv` `lbv` `stp`
p͡f		`vls` `bld` `afr`
b͡v		`mdv` `bld` `afr`

Gemination

Gemination is found in several languages including Italian and Japanese. It is also present in the suprasegmental phonology between words such as “lamppost” and “evenness”.

Some linguists use the long suprasegmental for geminate consonants. The eSpeak NG convention is to use consonant length for phonation when consonant length is distinct without gemination occurring.

The way gemination is represented in eSpeak NG is to duplicate the phonemes, with the first phoneme using the unx feature. For example, n̚.n for a geminated n. This describes how with the stp and nas consonants, the mouth remains closed (unx) for the first of the geminated consonants.

Vowels

	`fnt`		`cnt`		`bck`
`hgh`	i	y	ɨ	ʉ	ɯ	u
`smh`	ɪ	ʏ				ʊ
`umd`	e	ø	ɘ	ɵ	ɤ	o
`mid`			ə
`lmd`	ɛ	œ	ɜ	ɞ	ʌ	ɔ
`sml`	æ		ɐ
`low`	a	ɶ			ɑ	ɒ

Symbols to the left are unr, and to the right are rnd.

NOTE: The smh vowels are more cnt than the other vowels. However, this distinction is not needed to classify these vowels, so is not included in the above table.

Positioning Diacritics

The following IPA diacritics are only used by eSpeak NG to fill out positions in the IPA consonant and vowel charts. As such those phonemes are transcribed according to the features at that position, not using the features at the location of the base phoneme with a feature for each of the positioning diacritics.

Symbol	Name
◌̟	advanced
◌̠	retracted
◌̈	centralized
◌̽	mid-centralized
◌̝	raised
◌̞	lowered

Features

Manner of Articulation

The manner of articulation is described in terms of several distinct feature types. The possible manners of articulation are:

Manner of Articulation	Feature	Symbol	Features
nasal	`nas`		`pmc egs nsl occ`
plosive (stop)	`stp`		`pmc egs orl occ`
affricate	`afr`		`pmc egs orl occ frr`
fricative	`frc`		`pmc egs orl frv`
tap/flap	`flp`		`pmc egs orl fla`
trill	`trl`		`pmc egs orl tri`
approximant	`apr`		`pmc egs orl app`
click	`clk`		`vlc igs orl`
ejective	`ejc`		`vlc igs orl occ`
implosive	`imp`	◌ʼ	`gtc igs`
vowel	`vwl`		`pmc egs orl vow`

The features for these manners of articulation are provided for convenience, and to make it easier to describe the IPA consonants. Internally, the distinct feature types are used.

For imp consonants, they use the features of the base phoneme except for the pmc and egs features. Thus, a nas imp is a gtc igs nsl occ.

The vwl phonemes are described using vowel height and backness features, while consonants (the other manners of articulation) are described using place of articulation features.

Additionally, the manner of articulation can be refined using the following features:

Feature	Name	Description
`lat`	lateral	The air flow is directed along the sides of the tongue.
`sib`	sibilant	The air flow is directed through the teeth with the tongue.

Air Flow

Feature	Symbol	Name	Description
`egs`	↑	egressive	The air flow is moving outwards from the initiator to the target.
`igs`	↓	ingressive	The air flow is moving inwards from the target to the initiator.

The ↑ and ↓ symbols are from the extended IPA^[7]. They only need to be used when the air flow is different to the base IPA phoneme (e.g. using ↓ on pulmonic consonants).

Initiator

Feature	Name	Description
`pmc`	pulmonic	The diaphragm and lungs are used to generate the airstream.
`gtc`	glottalic	The glottis is used to generate the airstream.
`vlc`	velaric	The velum is closed and the tongue is used to generate the airstream.
`pcv`	percussive	There is no airstream used to produce this sound.

Target

Feature	Name	Description
`nsl`	nasal	The air flows through the nose.
`orl`	oral	The air flows through the mouth.

Manner

Feature	Name	Description
`occ`	occlusive	The air flow is blocked within the vocal tract.
`frv`	fricative	The air flow is constricted, causing turbulence.
`fla`	flap	A single tap of the tongue against the secondary articulator.
`tri`	trill	A rapid vibration of the primary articulator against the secondary articulator.
`app`	approximant	The vocal tract is narrowed at the place of articulation without being turbulant.
`vow`	vowel	The phoneme is articulated as a vowel instead of a consonant.

Phonation

The phonation features describe the degree to which the glottis (vocal chords) are open or closed.

Feature	Symbol	Name	Description
`vls`		voiceless	The glottis is fully open, such that the vocal chords do not vibrate.
`brv`	◌̤	breathy voice	The glottis is closed slightly, to produce a whispered or murmured sound.
`slv`	◌̥	slack voice	The glottis is opened wider than `mdv`, but not enough to be `brv`.
`mdv`		modal voice	The glottis is opened to provide the optimal vibration of the vocal chords.
`stv`	◌̬	stiff voice	The glottis is closed narrower than `mdv`, but not enough to be `crv`.
`crv`	◌̰	creaky voice	The glottis is closed to produce a vocal or glottal fry.
`glc`	ʔ͡◌	glottal closure	The glottis is fully closed.

The IPA ◌̥ diacritic is also used to fill the vls spaces in the IPA consonant charts. Thus, when ◌̥ is used with a mdv consonant that does not have an equivalent vls consonant, the resulting consonant is vls, not slv.

Place of Articulation

The place of articulation is described in terms of an active articulator and one or more passive articulators^[9]. The possible places of articulation are:

Place of Articulation	Feature	Symbol	Active	Lips	Teeth	Passive
bilabial	`blb`		`lbl`	`ulp`
linguolabial	`lgl`	◌̼	`lmn`	`ulp`
labiodental	`lbd`	◌̪	`lbl`		`utt`
bilabial-labiodental	`bld`		`bld`	`ulp`	`utt`
interdental	`idt`	◌̪͆	`lmn`		`utt`
dental	`dnt`	◌̪	`apc`		`utt`
denti-alveolar	`dta`		`lmn`		`utt`	`alf`
alveolar	`alv`		`lmn`			`alf`
apico-alveolar	`apa`	◌̺	`apc`			`alf`
palato-alveolar	`pla`		`lmn`			`alb`
apical retroflex	`arf`		`sac`			`alb`
retroflex	`rfx`	◌̺	`apc`			`hpl`
alveolo-palatal	`alp`		`dsl`			`alb`
palatal	`pal`		`dsl`			`hpl`
velar	`vel`		`dsl`			`spl`
labio-velar	`lbv`		`dsl`	`ulp`		`spl`
uvular	`uvl`		`dsl`			`uvu`
pharyngeal	`phr`		`rdl`			`prx`
epiglotto-pharyngeal	`epp`		`lyx`			`prx`
(ary-)epiglottal	`epg`		`lyx`			`egs`
glottal	`glt`		`lyx`			`gts`

The features for these places of articulation are provided for convenience, and to make it easier to describe the IPA consonants. Internally, the active and passive articulators are used.

The ◌̪ diacritic is lbd when used on blb consonants, and dnt when used on alv consonants.

The bld place of articulation is used for afr consonants that have a blb onset and a lbd release, e.g. in the p͡f consonant.

The alv consonant is lmn as found in French and Spanish, while apa is apc as found in English, as such ◌̻ (laminal) is not needed.

NOTE: The IPA charts make a distinction between pharyngeal and epiglottal consonants, but Wikipedia does not. This model uses the Wikipedia descriptions.

Active Articulators

Feature	Name	Articulator
`lbl`	labial	lower lip
`lmn`	laminal	tongue blade
`apc`	apical	tongue tip
`sac`	subapical	underside of the tongue
`dsl`	dorsal	tongue body
`rdl`	radical	tongue root
`lyx`	laryngeal	larynx

Passive Articulators

Feature	Articulator
`ulp`	upper lip
`utt`	upper teeth
`alf`	alveolar ridge (front)
`alb`	alveolar ridge (back)
`hpl`	hard palate
`spl`	soft palate (velum)
`uvu`	uvular
`prx`	pharynx
`egs`	epiglottis
`gts`	glottis

Rounding and Labialization

Feature	Symbol	Name	Rounded	Position
`unr`		unrounded	No	Close to the jaw.
`ptr`	◌ʷ, ◌ᶣ	protruded	Yes	Protrude outward from the jaw.
`cmp`	◌ᵝ	compressed	Yes	Close to the jaw.
`rnd`		rounded	Yes	`ptr` if `bck` or `cnt`; `cmp` if `fnt`

The unr and rnd features are used for vowels to describe their default labialization. Consonants are unr by default, and can use the ◌ʷ, ◌ᶣ and ◌ᵝ annotations to specify the type of labialization. Vowels can use these to change their labialization from the default one specified by rnd.

Additionally, the degree of rounding/labialization can be specified using the following features:

Feature	Symbol	Name
`mrd`	◌̹	more rounded
`lrd`	◌̜	less rounded

Vowel Height

Feature	Name
`hgh`	close (high)
`smh`	near-close (semi-high)
`umd`	close-mid (upper-mid)
`mid`	mid
`lmd`	open-mid (lower-mid)
`sml`	near-open (semi-low)
`low`	open (low)

Vowel Backness

Feature	Name
`fnt`	front
`cnt`	center
`bck`	back

Syllabicity

Feature	Symbol	Name
`syl`	◌̩	syllabic
`nsy`	◌̯	non-syllabic

Consonant Release

Feature	Symbol	Name
`frr`		fricative release
`asp`	◌ʰ	aspirated
`nrs`	◌ⁿ	nasal release
`lrs`	◌ˡ	lateral release
`unx`	◌̚	no audible release (unexploded)

Co-articulation

Feature	Symbol	Name	Co-Articulator	Type
`pzd`	◌ʲ	palatalized	`hpl`	Passive Articulator
`vzd`	◌ˠ	velarized	`spl`	Passive Articulator
`fzd`	◌ˤ	pharyngealized	`prx`	Passive Articulator
`nzd`	◌̃	nasalized	`nsl`	Target
`rzd`	◌˞	rhoticized	`rfx`	Place of Articulation

Additionally, the tongue root position can be specified using the following features:

Feature	Symbol	Name
`atr`	◌̘	advanced tongue root
`rtr`	◌̙	retracted tongue root

Fortis and Lenis

Feature	Symbol	Name
`fts`	◌͈	fortis
`lns`	◌͉	lenis

The extended IPA^[7] ◌͈ and ◌͉ diacritics are used to specify lesser (lns) and greater (fts) oral pressure than the unmodified voiced or voiceless phoneme. This distinction is made by the Ewe, Tabasaran, Archi, and other languages^[8].

Where fortis and lenis are used to contrast consonant durations (e.g. in the Jawoyn, Ojibwe, and Zurich German languages^[8]), the length suprasegmentals are used instead.

Suprasegmentals

Stress

Feature	Symbol	Name
`st1`	ˈ◌	primary stress
`st2`	ˌ◌	secondary stress
`st3`	ˈˈ◌	extra stress

Length

Feature	Symbol	Name
`est`	◌̆	extra short
`hlg`	◌ˑ	half-long
`lng`	◌ː	long

Rhythm

Feature	Symbol	Name
`sbr`	◌.◌	syllable break
`lnk`	◌‿◌	linked (no break)

Intonation

Feature	Symbol	Name
`fbr`	\|	minor (foot) break
`ibr`	‖	major (intonation) break
`glr`	↗	global rise
`glf`	↘	global fall

Tone Stepping

Feature	Symbol	Name
`ust`	ꜛ◌	upstep
`dst`	ꜜ◌	downstep

Properties

Tones

Tones are defined using the following 3 properties:

tone_start  <value>
tone_middle <value>
tone_end    <value>

The <value> field for these properties is a number with one of the following values:

Tone	Symbol	`<value>`
extra high (top)	◌˥	`5`
high	◌˦	`4`
mid	◌˧	`3`
low	◌˨	`2`
extra low (bottom)	◌˩	`1`

A level tone can be specified by just using the tone_start value. A raising or falling tone can be specified using the tone_start and tone_end values. A raising-falling (peaking) or falling-raising (dipping) tone can be specified using all three values.

References

Kirshenbaum, Evan, Representing IPA phonetics in ASCII (HTML). 1993.
Kirshenbaum, Evan, Representing IPA phonetics in ASCII (PDF). 2001.
International Phonetic Association, The International Phonetic Alphabet and the IPA Chart. 2015. Creative Commons Attribution-Sharealike 3.0 Unported License (CC-BY-SA).
Wikipedia. International Phonetic Alphabet. 2017. Creative Commons Attribution-Sharealike 3.0 Unported License (CC-BY-SA).
Dunn, R. H., Cainteoir Text-to-Speech Phoneme Features. 2013-2015.
Wikipedia. Voiced glottal fricative. 2017, Creative Commons Attribution-Sharealike 3.0 Unported License (CC-BY-SA).
Wikipedia. Extensions to the International Phonetic Alphabet. 2017, Creative Commons Attribution-Sharealike 3.0 Unported License (CC-BY-SA).
Wikipedia. Fortis and lenis. 2017, Creative Commons Attribution-Sharealike 3.0 Unported License (CC-BY-SA).
Wikipedia. Place of articulation. 2017, Creative Commons Attribution-Sharealike 3.0 Unported License (CC-BY-SA).

phonemes.md 31KB Raw Blame History