How many singing genres are actually distinguishable?

153 solo-voice styles, 5 takes each, 765 clips — every one sung to the same lyrics so that any measured difference is the singing and not the words.

the answer

~50

distinguishable genres, not 153

clips

765

153 genres × 5, 0 failures

best view

VoiceCLAP

between/within ratio 1.67

music embedding

1.00

ratio — no signal at all

First, a dead end worth publishing

The obvious embedding sees nothing. music-whisper — the 23040-d encoder the aesthetics and popularity heads are built on, and the one anybody would reach for — gives a between-genre centroid distance of 1.0038 against a within-genre distance of 1.0048. A ratio of 1.00. Leave-one-out accuracy over 153 genres is 0.004, where chance is 0.0065 — that is chance. Every genre's margin to its nearest neighbour is negative, and the confusions it does make are nonsense: dramatic soprano → Carnatic, countertenor → vaporwave croon.

The tempting headline from that number is "153 singing genres are indistinguishable". It would have been wrong, and the check that shows it is cheap: measure the audio directly.

genre	intended	median F0
playground chant	child	344 Hz
child folk singer	child	325 Hz
teen pop	teen	279 Hz
boy treble	child	232 Hz
basso profondo	older male	177 Hz
Tuvan throat singing	adult male	163 Hz
blues delta	older male	157 Hz
elderly folk singer	older male	148 Hz

F0 spans 139–350 Hz, ordered exactly as the age and register targets specify, and hardcore_scream is the noisiest clip in the set by both spectral flatness and centroid. The generator did render the genres differently. The encoder simply cannot represent it. music-whisper is trained on full mixtures for aesthetics and popularity; on dry solo voice with identical lyrics it is near-isotropic. Anyone scaling this work would reach for it by default, so it is worth knowing it cannot do this job.

Three views that can see it

Two representations built for voice, plus a deliberately simple hand-built acoustic vector as a control. Lift is accuracy divided by chance — raw accuracy over 153 classes is meaningless on its own.

representation	dims	between	within	ratio	LOO acc	lift vs chance
music-whisper	23040	1.0038	1.0048	1.00	0.004	0.6×
VoiceCLAP	768	0.9752	0.5838	1.670	0.035	5.4×
VoiceNet dimensions	57	0.9718	0.6251	1.555	0.047	7.2×
acoustic, hand-built	42	0.9979	0.7858	1.270	0.014	2.2×

A 42-dimensional hand-built feature vector beats a 23040-d learned embedding by 550× on lift. That is not a subtle margin. For solo voice, pitch statistics, dynamics, spectral shape and MFCCs carry the genre signal, and a music-mixture encoder throws it away. If you scale this, pick the representation on evidence rather than on parameter count.

Where genres stop being separable

Merge the 153 genre centroids agglomeratively and re-run the probe at each level. Lift peaks where the number of clusters matches the number of things the model actually renders differently.

clusters	VoiceCLAP lift	VoiceNet dimensions lift	acoustic, hand-built lift
153	5.4×	7.2×	2.2×
100	11.1×	11.5×	10.8×
80	12.4×	12.2×	12.4×
64	11.7×	12.4×	12.1×
50	12.9×	11.4×	12.7×
40	11.2×	10.2×	11.1×
32	10.5×	9.5×	9.6×
24	9.0×	8.3×	8.5×
16	8.4×	6.5×	7.1×
12	7.1×	5.7×	5.9×
8	5.8×	5.0×	4.9×
6	4.6×	4.0×	3.9×
4	3.4×	3.1×	3.1×

The answer: roughly 50–80, not 153.
All three independent views peak in the same band — VoiceCLAP and the acoustic baseline at 50, VoiceNet at 64 — and the plateau is flat from about 50 to 100 (lift 10.9–12.9×). Above 100 it falls away, and at the full 153 it collapses to 2.2–7.2×. Three representations built on different principles agreeing on the same band is the strongest evidence available here.

For generating training data, ~64 categories is the defensible target. Keeping all 153 is not harmful — the extra labels are just finer than the model can reliably realise, so they add annotation precision the audio does not support.

The families are wrong, and that is useful

Nearest-neighbour confusions stay inside my hand-assigned family only 28/153 of the time (VoiceCLAP; the other views agree at 30/153 and 38/153). If the taxonomy matched the acoustics, that number would be high.

It is not a bug in the measurement — it says the way these styles are conventionally grouped (pop, rock, world, classical) is not how they actually sound when stripped to one dry unaccompanied voice. A Carnatic line and a dramatic soprano share far more, acoustically, than a dramatic soprano and a Gregorian chant. If you want the clusters to be meaningful, derive them from the audio rather than from genre history.

Can it sing as a child? Sometimes — and the failures have a pattern

Twelve entries were capability probes rather than assumptions. Measured with VoiceNet's age dimension (AGEV, higher = older):

intended age band	genres	mean AGEV
child	5	2.022
teen	5	2.356
young adult	26	2.453
adult	108	3.117
older	9	3.181

The ordering is monotonic and correct — the age instruction does work on average. But per genre it is bimodal, and which way it falls is predictable:

probe	intended	AGEV	outcome
child folk singer	child	1.133	landed
playground chant	child	1.198	landed
teenager singing alone	teen	1.233	landed
teen pop	teen	1.348	landed
nursery rhyme, child	child	1.444	landed
elderly folk singer	older	3.652	landed
lullaby, grandmother	older	1.458	came out young
emo	teen	3.006	adult
boy treble	child	3.089	adult
adolescent solo	teen	3.091	adult
boy-band tenor	teen	3.100	adult
chorister solo	child	3.248	adult

Age and style compete in the caption, and style wins. Six of twelve probes landed. Every one that landed had an age-neutral or child-associated style clause — nursery rhyme, playground, bedroom. Every one that failed carried a strong adult stylistic identity — chorister, boy-band, emo, grandmother's lullaby — and the model rendered the idiom and discarded the age.

So yes, the model can sing as a child, reliably, when nothing in the prompt pulls the other way. To get a child chorister the age has to be reinforced against the style, not merely stated alongside it.

Listen — one take per genre

Seed 1234, lyric set A, 160 kbps stereo. AGEV is VoiceNet's age reading of that genre's takes.

blues — 4 genres

genre	intended voice	AGEV	style clause
Chicago blues shouter blues_chicago	adult · male	3.98	chicago blues shouter, powerful gravelly belt, urgent and declamatory
Classic blues blues_classic_female	adult · female	2.63	classic 1920s blues woman, broad vibrato, sly slow drawl, theatrical
Delta blues blues_delta	older · male	4.14	delta blues, rough weathered voice, bent notes, moaning field-holler inflection
Field holler field_holler	adult · male	3.74	unaccompanied field holler, long falling cries, free rhythm, raw and open-throated

classical — 16 genres

genre	intended voice	AGEV	style clause
Baroque ornamented aria baroque_ornament	adult · female	2.24	baroque aria with period ornamentation, straight tone, trills and appoggiaturas
Basso profondo opera_basso	older · male	3.75	basso profondo, extremely low sonorous voice, cavernous resonance
Bel canto bel_canto	adult · any	3.00	bel canto, effortless even legato across registers, elegant messa di voce swells
Boy treble / chorister (probe) boy_treble	child · male	3.09	boy treble chorister, pure clear unbroken child voice, straight tone, cathedral style
Byzantine chant byzantine	adult · male	3.43	byzantine chant, nasal modal melisma, microtonal inflection, liturgical
Chorister solo (probe) choirboy_solo	child · male	3.25	solo chorister boy, pure vibrato-free treble, echoing cathedral clarity
Coloratura soprano opera_sop_coloratura	adult · female	1.78	coloratura soprano, very high tessitura, rapid agile ornament runs, bright ringing tone
Countertenor countertenor	adult · male	3.55	countertenor, male falsetto in the alto range, pure straight early-music tone
Dramatic soprano aria opera_sop_dram	adult · female	2.58	dramatic operatic soprano, full chest-supported legato, wide slow vibrato, Italianate aria
German Lied lieder	adult · any	3.51	art-song lieder delivery, intimate recital voice, careful diction, restrained vibrato
Gregorian chant gregorian	adult · male	3.56	solo gregorian plainchant, free unmetred modal line, monastic and unaccompanied
Lyric tenor opera_tenor_lirico	adult · male	3.40	lyric operatic tenor, ringing squillo, open Italianate vowels, soaring line
Mezzo-soprano opera_mezzo	adult · female	2.17	mezzo-soprano, warm dark middle register, rich operatic legato
Operatic baritone opera_baritone	adult · male	3.82	operatic baritone, dark resonant chest voice, noble sustained line
Oratorio / sacred solo oratorio	adult · any	3.55	oratorio soloist, clean sustained sacred line, minimal vibrato, devotional restraint
Sprechgesang sprechgesang	adult · any	3.55	sprechgesang, half-spoken half-sung expressionist delivery, pitched speech

country — 6 genres

genre	intended voice	AGEV	style clause
Bluegrass gospel gospel_bluegrass	adult · male	3.48	bluegrass gospel, hard-driving high tenor, fervent and sharp
Classic country country_classic	adult · male	3.46	classic country, pronounced twang, cry-break yodel catch in the voice
Country pop country_pop	young_adult · female	1.73	country pop, bright polished twang, conversational verses, big clean chorus
Cowboy yodel cowboy_yodel	adult · male	3.46	cowboy yodelling, rapid chest-to-falsetto breaks, open prairie tone
High lonesome bluegrass bluegrass_high_lonesome	adult · male	3.36	high lonesome bluegrass tenor, piercing nasal keening, hard mountain tone
Outlaw country outlaw_country	older · male	3.51	outlaw country, weathered low drawl, gruff and unhurried

electronic — 8 genres

genre	intended voice	AGEV	style clause
Ambient wordless vocal ambient_wordless	adult · female	1.85	wordless ambient vocal, floating sustained vowel tones, no consonants
Drum and bass vocal dnb_vocal	young_adult · female	1.85	liquid drum and bass vocal, soulful clipped phrases, urgent and bright
House diva house_diva	adult · female	2.20	house diva, gospel-derived belt, euphoric sustained hooks
Hyperpop hyperpop	young_adult · any	1.50	hyperpop, extreme pitched-up processed brightness, hyperactive delivery
Industrial industrial_shout	adult · male	3.82	industrial vocal, distorted barked mechanical shouting
Synthwave synthwave	adult · male	3.41	synthwave, breathy nostalgic eighties-styled tenor, reverbless close delivery
Trance vocal trance_vocal	young_adult · female	1.84	trance vocal, airy ethereal sustained high line, wide open vowels
Vaporwave croon vaporwave_croon	adult · male	3.58	slowed vaporwave croon, syrupy pitched-down lounge delivery

folk — 5 genres

genre	intended voice	AGEV	style clause
Appalachian ballad appalachian_ballad	older · female	2.31	unaccompanied appalachian ballad, plain hard modal tone, ornamented free rhythm
Sea shanty sea_shanty	adult · male	3.44	sea shanty, hearty rolling call, blunt communal projection
Shape-note / Sacred Harp shape_note	adult · any	3.25	shape-note sacred harp singing, hard open unpolished tone, fierce plain delivery
Singer-songwriter americana_singer_songwriter	adult · any	3.53	singer-songwriter, plainspoken confessional delivery, small honest untrained voice
Work song work_song	adult · male	3.35	call-and-response work song, steady labouring pulse, blunt strong voice

functional — 18 genres

genre	intended voice	AGEV	style clause
Adolescent solo (probe) teen_choir_solo	teen · male	3.09	adolescent boy whose voice is breaking, unstable cracking register shifts
Auctioneer chant auctioneer	adult · male	3.34	auctioneer chant, extremely fast rhythmic filler-syllable patter
Beatboxing beatbox	young_adult · male	3.42	beatboxing, percussive mouth drum sounds, rhythmic and unpitched
Child folk singer (probe) child_folk	child · female	1.13	young girl singing a folk tune, clear small unaffected child voice
Elderly folk singer (probe) elderly_folk	older · male	3.65	very old man singing, wavering unsteady pitch, thin breathy worn voice
Humming humming	adult · any	3.71	closed-mouth humming, no words, warm resonant nasal tone
Keening / lament keening	older · female	2.19	ritual keening lament, sobbing wailing cries, grief-stricken and unmetred
Long-form modal call muezzin_style	adult · male	3.19	long-form unaccompanied modal call, sustained ornamented open-air projection
Lullaby lullaby	adult · female	2.03	lullaby, extremely soft rocking repetitive line, hushed and tender
Lullaby, older voice (probe) lullaby_grandmother	older · female	1.46	lullaby sung by an old woman, thin frail wavering tone, worn and gentle
Maximum belt belt_max	adult · female	2.36	maximum-effort belting, full-throated sustained loud chest voice, strained power
Nursery rhyme, child (probe) nursery_rhyme_child	child · any	1.44	young child singing a nursery rhyme, small light unformed voice, uneven pitch, artless
Playground chant (probe) playground_chant	child · female	1.20	children's playground chant, sing-song taunting cadence, shouted and rhythmic
Pure falsetto falsetto_pure	adult · male	3.38	pure falsetto, light hooty unsupported head voice, fragile
Teenager singing alone (probe) teen_bedroom	teen · female	1.23	teenager singing alone in a bedroom, untrained thin voice, self-conscious
Vocal fry singing vocal_fry_sing	young_adult · female	1.50	vocal fry singing, creaky low rattling register, lazy and textural
Whisper singing whisper_sing	adult · female	1.60	whisper singing, almost no vocal fold tone, breath-dominant intimate
Whistling whistling	adult · any	3.63	solo whistling a melody, pure clear tone, no voice

jazz — 5 genres

genre	intended voice	AGEV	style clause
Crooner jazz_crooner	adult · male	3.66	mid-century crooner, velvet baritone, close-mic intimacy, effortless swing
Jazz standard jazz_standard	adult · female	2.62	jazz standard vocal, behind-the-beat phrasing, warm smoky tone, subtle swing
Scat singing jazz_scat	adult · any	4.00	scat singing, improvised nonsense syllables, instrumental bebop agility
Torch song torch_song	adult · female	2.33	torch song, aching slow ballad, dramatic rubato, world-weary and confessional
Vocalese vocalese	adult · any	3.43	vocalese, lyrics set to a fast instrumental solo line, dense rapid articulation

metal — 5 genres

genre	intended voice	AGEV	style clause
Black metal shriek metal_black	adult · male	3.71	black metal shriek, high rasping strangled screech
Clean metal baritone metal_clean_baritone	adult · male	3.96	clean metal baritone, dark brooding sustained tone, gothic gravity
Death growl metal_death	adult · male	3.88	death metal growl, guttural subharmonic roar, unpitched and cavernous
Gothic metal soprano metal_gothic_female	adult · female	2.33	gothic symphonic metal soprano, operatic vibrato over heaviness, ethereal
Power metal metal_power	adult · male	3.32	power metal, soaring operatic high tenor, sustained ringing screams

pop — 13 genres

genre	intended voice	AGEV	style clause
80s synthpop synthpop_80s	adult · male	3.30	1980s synthpop, theatrical clipped diction, cool detached delivery
Bedroom pop bedroom_pop	young_adult · female	1.51	bedroom pop, whispered soft double-tracked intimacy, unforced small voice
Boy-band tenor (probe) pop_boyband	teen · male	3.10	boy-band lead, light high tenor, earnest and smooth, adolescent brightness
Eurodance diva europop_dance	young_adult · female	2.12	eurodance diva, powerful sustained belt over four-to-the-floor energy
French chanson chanson	adult · female	2.78	french chanson, dramatic rolled consonants, speech-like emotive rubato
Indie pop indie_pop	young_adult · any	2.63	indie pop, slightly nasal untrained charm, close-mic breathiness, casual
Italian canzone canzone	adult · male	3.50	italian canzone, big open-throated romantic delivery, soaring passionate line
J-pop jpop	young_adult · female	1.21	j-pop, high clear cute-inflected tone, wide melodic leaps, earnest
Japanese city pop citypop	young_adult · female	1.46	japanese city pop, bright polished eighties tone, breezy sophisticated phrasing
K-pop kpop	young_adult · female	1.42	k-pop idol vocal, crisp bright tone, precise runs, glossy modern production style
Pop ballad pop_ballad	young_adult · female	2.06	modern pop ballad, clean controlled belt, emotional build to a big chorus
Schlager schlager	adult · female	1.98	german schlager, warm sentimental major-key delivery, wide friendly vibrato
Teen pop (probe) pop_teen	teen · female	1.35	teen pop, bright youthful thin-bodied voice, eager and unweathered

rock — 8 genres

genre	intended voice	AGEV	style clause
Britpop britpop	young_adult · male	3.21	britpop, adenoidal regional-accented drawl, cocky flat delivery
Classic rock rock_classic	adult · male	4.18	classic rock, raspy powerful high belt, bluesy bends, arena projection
Emo (probe) emo	teen · male	3.01	emo, cracking earnest voice, whiny nasal edge tipping into desperate yells
Glam rock rock_glam	adult · male	3.73	glam rock, theatrical androgynous sneer, camp vibrato
Grunge grunge	young_adult · male	3.64	grunge, strained throaty rasp, apathetic verses erupting into raw shouting
Hardcore scream hardcore_scream	young_adult · male	3.32	hardcore screaming, throat-shredding roars, barely pitched
Post-rock wail post_rock_wail	adult · any	3.78	wordless post-rock wailing, distant reverberant keening, textural
Punk punk	young_adult · male	3.54	punk, shouted snarling barked delivery, aggressive and untrained

soul — 7 genres

genre	intended voice	AGEV	style clause
Classic soul soul_classic	adult · male	4.05	1960s soul, gritty impassioned voice, raw cries and pleading edge
Contemporary R&B rnb_contemporary	young_adult · female	2.06	modern R&B, airy layered runs, melismatic ad-libs, polished intimate delivery
Doo-wop lead doo_wop	young_adult · male	3.19	doo-wop lead, sweet earnest tenor, swooping falsetto leaps, vintage warmth
Gospel lead gospel_lead	adult · female	2.25	gospel lead, huge melismatic runs, church belt, ecstatic and rising
Neo-soul neo_soul	young_adult · female	1.53	neo-soul, breathy conversational phrasing, jazzy runs, laid-back pocket
Quiet storm quiet_storm	adult · male	3.53	quiet storm R&B, smooth low seductive croon, silky sustained falsetto
Spiritual spiritual	adult · any	3.60	unaccompanied spiritual, slow dignified line, deep sorrowful weight

theatre — 5 genres

genre	intended voice	AGEV	style clause
Cabaret cabaret	adult · any	3.15	cabaret, arch knowing delivery, spoken asides sliding into sung line
Musical theatre belt musical_belt	adult · female	2.02	musical theatre belt, forward bright chest mix, crystal diction, big finish
Musical theatre legit musical_legit	adult · female	1.94	legit musical theatre soprano, light lifted classical-adjacent tone
Patter song patter_song	adult · male	3.71	comic patter song, extremely fast precise consonants, breathless wit
Vaudeville vaudeville	adult · male	3.79	vaudeville, brassy nasal old-time projection, exaggerated diction

urban — 8 genres

genre	intended voice	AGEV	style clause
Afrobeats afrobeats	young_adult · male	2.96	afrobeats, lilting melodic pidgin-inflected phrasing, relaxed buoyant swing
Dancehall dancehall	adult · male	3.35	dancehall, rapid patois toasting sliding into melody, rhythmic and percussive
Drill hook uk_drill_sung	young_adult · male	3.81	drill hook, deadpan low sung-spoken menace, clipped london phrasing
Grime hook grime_hook	young_adult · male	3.31	grime hook, shouted urgent london delivery, high energy
Lovers rock lovers_rock	young_adult · female	2.10	lovers rock, sweet gentle reggae-soul, tender high register
Melodic rap melodic_rap	young_adult · male	3.40	melodic rap singing, half-sung half-spoken, autotune-adjacent slides
Roots reggae reggae_roots	adult · male	3.83	roots reggae, nasal plaintive tone, off-beat phrasing, prophetic delivery
Sung trap hook trap_sung	young_adult · male	3.48	sung trap hook, repetitive catchy melodic phrase, lazy drawled delivery

world — 45 genres

genre	intended voice	AGEV	style clause
Andean huayno andean_huayno	adult · female	1.89	andean huayno, very high piercing nasal quechua-inflected tone
Arabic mawwal arabic_mawwal	adult · male	3.58	arabic mawwal, free-rhythm maqam improvisation, long ornamented tarab phrases
Balkan open-throat balkan_open	adult · female	2.21	balkan open-throat singing, piercing straight-tone chest voice, sharp dissonant edge
Bhajan raga_bhajan	adult · female	2.05	devotional bhajan, simple repetitive ecstatic line, warm and communal
Bolero bolero_latin	adult · male	4.15	latin bolero, velvet romantic croon, dramatic sustained romantic line
Bossa nova bossa_nova	adult · female	2.16	bossa nova, breathy near-whispered understatement, gentle behind-beat
Bulgarian diaphonic bulgarian_diaphonic	adult · female	1.96	bulgarian village singing, bright nasal forward tone, tight ornament shakes
Cantonese opera cantonese_opera	adult · female	2.33	cantonese opera, extremely high nasal stylised tone, sharp stylised declamation
Cantorial cantorial	adult · male	3.44	cantorial chazzanut, virtuosic ornamented liturgical cries, huge emotive range
Carnatic kriti carnatic	adult · female	1.84	carnatic vocal, rapid gamaka oscillations, precise ornamented south indian line
Cumbia cumbia	adult · female	1.92	cumbia, lilting cheerful mid-range delivery, steady danceable phrasing
Enka enka	adult · female	2.21	japanese enka, heavy kobushi vibrato bends, melodramatic sentimental
Flamenco cante jondo flamenco_cante	adult · male	3.48	flamenco cante jondo, raw cracked melismatic wailing, microtonal anguish
Forró forro	adult · male	3.29	forro, warm northeastern brazilian accent, rustic bouncing delivery
Georgian solo line georgian_solo	adult · male	3.88	georgian traditional solo line, deep modal chant-like voice
Ghazal ghazal	adult · male	3.78	ghazal, poetic urdu delivery, tender ornamented restraint
Hawaiian leo kiʻekiʻe hawaiian_falsetto	adult · male	3.63	hawaiian falsetto, deliberate audible register breaks, sweet gentle tone
Hindustani khayal khayal	adult · male	3.62	hindustani khayal, slow raga exposition, microtonal meend glides, sustained alap
Irish sean-nós sean_nos	older · male	3.83	irish sean-nos, unaccompanied highly ornamented free-rhythm gaelic singing, nasal
Japanese min'yō min_yo	adult · female	2.05	japanese folk min-yo, tight nasal ornament, bright piercing tone
Katajjaq-style inuit_katajjaq	adult · female	1.75	inuit-style throat game vocalising, rhythmic voiced and unvoiced panting patterns
Klezmer-style krekhts klezmer_krekhts	adult · male	3.61	klezmer-inflected singing, sobbing krekhts catches, ornamented yiddish lament
Korean pansori pansori	adult · female	1.97	korean pansori, rough husky forced tone, dramatic epic storytelling cries
Mongolian long song mongolian_urtiin	adult · male	3.34	mongolian urtiin duu long song, immense sustained ornamented high line
Māori waiata maori_waiata	adult · any	3.48	maori waiata, resonant communal chant-song, strong open delivery
Noh utai noh_utai	older · male	3.78	japanese noh utai chanting, guttural constricted slow intoned delivery
Nordic kulning kulning	adult · female	2.45	scandinavian kulning herding call, extremely loud piercing high head voice, outdoor
Overtone singing throat_overtone	adult · any	3.13	overtone singing, sustained drone with clear whistling harmonic melody above
Peking opera peking_opera	adult · male	3.58	peking opera, piercing stylised falsetto-adjacent tone, sharp percussive syllables
Persian avaz persian_avaz	adult · male	3.40	persian avaz, tahrir vocal break ornament, plaintive modal free rhythm
Portuguese fado fado	adult · female	2.25	portuguese fado, saudade-laden dramatic swells, throaty tragic ornament
Qawwali qawwali	adult · male	3.60	qawwali solo, ecstatic rising improvised melisma, forceful devotional cry
Ranchera ranchera	adult · male	3.72	mexican ranchera, huge open-throated grito, proud sustained belt
Rebetiko rebetiko	adult · male	3.74	greek rebetiko, nasal ornamented modal lament, smoky and defiant
Saeta flamenco_sae	adult · female	2.30	spanish saeta, unaccompanied processional cry from a balcony, raw and piercing
Salsa sonero salsa_sonero	adult · male	3.35	salsa sonero, improvised rhythmic call phrases, bright nasal projection
Samba samba	adult · male	3.45	samba, buoyant percussive syncopated phrasing, open joyful tone
Sardinian cantu sardinian_cantu	adult · male	3.84	sardinian solo cantu a chiterra, strained high tense declamation
Sertanejo sertanejo	adult · male	3.62	brazilian sertanejo, nasal emotive twang, sentimental duet-style delivery
Sindhen gamelan_sindhen	adult · female	1.64	javanese sindhen, floating soft melismatic line, delicate and unhurried
Sámi joik joik	adult · any	3.85	sami joik, cyclical wordless vocable chant, guttural and hypnotic
Tango tango	adult · male	4.07	argentine tango, dramatic rubato declamation, bitter and theatrical
Turkish uzun hava turkish_uzun_hava	adult · male	3.56	turkish uzun hava, free-metre lament, heavy nasal ornament
Tuvan throat singing throat_tuvan	adult · male	3.73	tuvan khoomei throat singing, low fundamental drone with whistling overtone
Waulking song scots_gaelic_waulking	adult · female	2.05	scottish waulking song, rhythmic repetitive labour cadence, strong plain voice

Method, and what it cannot tell you

Same lyrics for every genre, two variants. If each genre had its own words, part of any separation would be lexical and the distinguishable-genre count would come out inflated.
Leave-one-out nearest-centroid over 5 takes per genre. With 5 samples the per-genre centroid is noisy; the merge curve is more trustworthy than any single genre's number.
Generation: ACE-Step 1.5 XL SFT with dcw_enabled=False. With the library default left on, every one of these 765 clips would have been unusable.
This measures what the generator renders and what these encoders see — not what a human listener can distinguish. A listening study could plausibly find more distinctions than 64, or fewer. No human has scored these clips.
Gemini has not judged these yet. Perceived age, perceived style and prompt-match are the next pass; the numbers here are embeddings only, which is the order the evidence should be built in.

765 clips · 153 genres × 5 takes · 0 generation failures · 25 s each · 48 kHz · ACE-Step 1.5 XL SFT, 50 steps, CFG 7.0, DCW off.