es.md

NOTICE

By using all OR ANY PART of this project you MUST follow our license at https://github.com/DYVAUX/nnsvs-english-support/blob/main/LICENSE.md This includes using parts of our work in yours. If you have questions about the license please contact us. We should not have to include this notice in our files but bad actors had forced our hand.

---

Cleaned up reference provided by subpum

Other:

Phoneme	X-SAMPA	Description
br		Inhale
exh		Exhale
axh		Voiced Exhale
pau		Silence
q	?	Glottal Stop
cl		Closure (Held Consonants)
vf		Vocal Fry
trash		Junk Phoneme

Vowels:

Phoneme	X-SAMPA	Explaination	Example
a	a	central open unrounded vowel	Ana
e	e	close-mid front unrounded vowel	Eva
i	i	close front unrounded vowel	Linda
o	o	close-mid back rounded vowel	Tope
u	u	close back rounded vowel	Uno

Semi-Vowels:

(Semi-vowels function as consonants and perform "diphthong-like" tasks.)

Phoneme	X-SAMPA	Explaination	Example
y	j	palatal approximant	Aire (a y r e), tierra (t y e rr a), sayo (s a y o)
w	w	labial-velar approximant	Pausa (p a w s a), fuego (f w e gh o)

Consonants:

Phoneme	X-SAMPA	Explanation	Example
b	b	voiced bilabial plosive	baño (b a ny o)
bh	B	voiced bilabial fricative	abaco (a bh a k o)
ch	tS	voiceless postalveolar affricate	choripan (ch o r i p a n)
d	d	voiced alveolar plosive	día (d i a)
dh	D	voiced dental fricative	andina (a n dh i n a)
f	f	voiceless labiodental fricative	farol (f a r o l)
g	g	voiced velar plosive	gato (g a t o)
gh	G	voiced velar fricative	agua (a gh w a)
k	k	voiceless velar plosive	casa (k a s a)
l	l	lateral alveolar approximant	lame (l a m e)
ll	L	voiced palatal lateral approximant	pollo (p o ll o)
m	m	bilabial nasal	madre (m a dh r e)
n	n	alveolar nasal	nadie (n a dh j e)
ny	J	palatal nasal	ñoquis (ny o k i s)
p	p	voiceless bilabial plosive	puerta (p w e r t a)
r	4	alveolar flap	larga (l a r gh a)
rr	r	alveolar trill	rara (rr a r a)
s	s	voiceless alveolar fricative	suena (s w e n a)
sh	S	voiceless postalveolar fricative	show (sh o w)
t	t	voiceless alveolar plosive	tema (t e m a)
x	x	voiceless velar fricative	jamón (x a m o n)
y	j\	voiced palatal fricative	yace (y a s e)
z	T	voiced alveolar fricative	zeta (z e t a)

Dialect Allophones:

READ THE NOTES for how to handle these. Most of these will never be used.

Phoneme	X-SAMPA	Explanation	Example	Dialect
ts	ts	voiceless alveolar sibilant affricate	chala (ts a l a)	Reference allophone for ch.
zh	Z	voiced postalveolar fricative	ellos (e zh o s)	Reference allophone for ll.
sh	S	voiceless postalveolar fricative	chala (sh a l a)	Reference allophone for ll/ch.
h	h	voiceless glottal fricative	obispo (o bv i h p o)	Reference allophone for s/x.
v	v	voiced labiodental fricative	afgano (a v gh a n o)	Reference allophone for f (before voiced consonants).
dj	J\	voiced palatal plosive	guía (dj i a)	Reference allophone for g (before some vowels).
ng	N	velar nasal	dingo (d i ng g o)	Reference allophone for n (before velar consonants).

Notes:

• sh and zh are sometimes used as allophones for ll. Still label it as ll for dictionary compatiblity. However, these phonmemes are useful for foreign words.

• Some speakers enounter "yeísmo", or the merging of "ll" and "y". These speakers can label "ll" as y instead to slightly reduce the amount of training data needed.

• Some speakers merge the hard and soft "b/d/g" sounds. These speakers can label all "b/d/g" as [b/d/g] and ignore the [bh/dh/gh] phonemes. Speakers without this merger can also drop [bh/dh/gh] to simplify the dataset, which will cause most "b/d/g” instances to sound softer (due to more instances of the soft sound in speech). Doing so also creates a slight risk that the model may use a hard "b/d/g" at an incorrect time.

• The y phoneme works as both a consonant AND semivowel. This is thanks to NNSVS being able to learn context.

• x covers all "h/soft J/aspirated s" sounds contextually. So only one phoneme is required. For speakers who want to differentiate, the h phoneme could be used along with the others. Note that this will increase the amount of data required and won't be supported in the dictionary.

• f is used for both devoiced "f" and voiced "v" sounds.

Chilean speaker notes:

• Upper-class speakers use ts instead of ch.
• Lower-class speakers use sh instead of ch.
• In most cases you should still label these as ch for dictionary compatiblity. Only differenciate if the speaker between ts/sh and ch (optional).