“One accent for Spain, one for Latin America” comes with a list of four differences. We put all four through 14 real speakers (11 from Spain, 2 from Mexico, 1 from Chile), and only one survives: the spelling z / ce / ci is [θ] in Spain and [s] in Latin America. The other three show no regional split in this corpus — and the numbers are on this page.
Two dots per row: amber = the segment spelled z / ce / ci, green = the same speaker's s. Dots apart = two different sounds in that mouth; dots together = merged. The ratio is on the right.
Same text, different regions. The number after each word is the centroid of the target fricative in that word.
This is not about changing your accent; it is about knowing what you are hearing.
The same letter group is [θ] in Spain (tongue tip between the teeth, air escaping through the gap) and [s] in Latin America. So the “th” in gracias is not your hearing failing — it is the standard Peninsular reading.
Both are standard. What sounds off is mixing: starting a word with [θ] and sliding into [s], or switching from day to day. Choose a region and keep it.
[θ] is not [f] and not a “lispy s”. Mandarin has no dental fricative, so the usual substitutes are [s] (tongue raised, air down the middle) or [f] (lower lip on the upper teeth). The only test: is the tongue tip visible between the teeth?
Three differences that are usually listed as regional markers do not separate regions in these 14 read-aloud speakers. Numbers beat a disclaimer.
Caribbean and Chilean Spanish are said to aspirate it to [h]. Not here: the three regions' medians are within a few hundred Hz, and the ranges overlap almost completely. Read-aloud style flattens exactly this difference.
ES 3454 Hz (n=406) · MX 3944 Hz (n=92) · CL 4177 Hz (n=30) | range ES 566–5622 / MX 746–5914 / CL 317–4994Argentina's “sheísmo” is the famous case. But the spread between individuals swamps it — within one region our speakers run from 583 to 2902 Hz.
ES 1491 Hz (n=143) · MX 1256 Hz (n=33) · CL 658 Hz (n=14) | range ES 448–4691 / MX 535–3557 / CL 490–3986Same story, and the smallest sample of all (six segments from Chile). Any conclusion would be cherry-picked from three people.
ES 2337 Hz (n=68) · MX 2530 Hz (n=14) · CL 2019 Hz (n=6) | range ES 1016–4192 / MX 1298–3358 / CL 886–4186Corpus: VoxForge Spanish (GPL), 14 speakers, 145 recordings, region self-reported by each speaker (Español España / Mexicano / Chile). Several people read the same public-domain text, which is what allows a same-sentence comparison.
Finding the segment: forced alignment (MMS_FA) gives phoneme boundaries; we measure the middle 60%, because both ends are transitions that drag in neighbouring vowels.
Classifying: by spelling, not by phoneme label. gruut emits θ for Latin American text too, so classifying by label would bake the conclusion into the premise. Letter ownership comes from the same DP the aligner uses (align_units); only z, ce, ci and s count, so México's x and the c in acción are excluded (167 segments dropped).
Sample size: 11 from Spain, 2 from Mexico, 1 from Chile. Three Latin American speakers is not a survey: it shows these three lack the contrast, not that the whole region does.
Known bias: z/c segments are an order of magnitude rarer than s segments (49 vs 871 in Spain), so each speaker's ratio rests on 2–5 segments. Every row divides one speaker by himself; the cross-speaker comparison is ratios, not absolute centroids.
scripts/24e_es_accents.py; counts and percentages are injected by the script, none typed by hand.