Dataset Viewer
Auto-converted to Parquet Duplicate
clip_id
int64
2
592
filename
stringlengths
7
7
speaker
stringclasses
4 values
domain
stringclasses
3 values
duration_s
float64
1.73
28.4
n_tokens
int64
2
53
reference_text
stringlengths
14
306
2
002.wav
Speaker 1
Play
3.306667
6
E chesto capisce tu: 'e denare!
3
003.wav
Speaker 1
Play
6.101333
11
E cu' 'e denare t'he accattato tutto chello ca he voluto!
4
004.wav
Speaker 1
Play
7.829333
15
Ma Filumena Marturano ha fatto correre essa a te! E currive senza ca te n'addunave.
5
005.wav
Speaker 1
Play
8.704
21
E ancora he 'a correre, ancora he 'a iettà 'o sango a capi comme se campa e se prucede 'a galantomo!
6
006.wav
Speaker 1
Play
4.288
5
O miédeco nun sapeva niente.
7
007.wav
Speaker 1
Play
3.904
9
Ce ha creduto pur'isso, e ce avev' 'a credere!
8
008.wav
Speaker 1
Play
8.896
14
Qualunque femmena, doppo vinticinc'anne che ha passato vicino a te, se mette in agonia.
9
009.wav
Speaker 1
Play
2.88
4
T'aggio fatto 'a serva!
10
010.wav
Speaker 1
Play
4.8
10
A serva ll'aggio fatta pè vinticinc'anne, e vuie 'o ssapite.
11
011.wav
Speaker 1
Play
5.290667
8
E maie ca t'avesse visto sottomessa, che ssaccio?
12
012.wav
Speaker 1
Play
2.986667
6
E avev' 'a chiagnere pe' te?
13
013.wav
Speaker 1
Play
2.901333
5
Era troppo bello 'o mobile.
14
014.wav
Speaker 1
Play
4.416
8
E quanno me vulive vedé 'e durmi, tu?
15
015.wav
Speaker 1
Play
4.842667
8
A strada d' 'a casa t' 'a scurdave.
16
016.wav
Speaker 1
Play
6.037333
14
E mmeglie feste, 'e meglie Natale me ll'aggio passate sola comm' a na cana.
17
017.wav
Speaker 1
Play
3.008
4
Saie quanno se chiagne?
18
018.wav
Speaker 1
Play
4.16
10
Quanno se cunosce 'o bbene e nun se pò avé!
19
019.wav
Speaker 1
Play
8.704
17
Ma Filumena Marturano bene nun ne cunosce... e quanno se cunosce sulo 'o mmale nun se chiagne.
20
020.wav
Speaker 1
Play
5.376
11
A suddisfazione 'e chiagnere, Filumena Marturano, nun l'ha pututa maie avé!
21
021.wav
Speaker 1
Play
4.330667
7
Comm' all'ultima femmena m' he trattato, sempe!
22
022.wav
Speaker 1
Play
9.557333
19
Ma mo, all'urdemo all'urdemo, a cinquantaduie anne, se retira cu' 'e fazzulette spuorche 'e russetto, ca me fanno schifo.
23
023.wav
Speaker 1
Play
3.264
7
A chella chi? .,. A chella chi?
24
024.wav
Speaker 1
Play
2.602667
4
Appriesso a chella schifosa!
25
025.wav
Speaker 1
Play
4.032
7
Che te cride ca nun l'avevo capito?
26
026.wav
Speaker 1
Play
4.288
12
Tu buscie nun ne saie dicere, e chisto è 'o difetto tuio.
27
027.wav
Speaker 1
Play
5.397333
13
Cinquantaduie anne, e se permette 'e se mettere cu' na figliola 'e vintiduie!
28
028.wav
Speaker 1
Play
2.368
5
Nun se ne mette scuorno
29
029.wav
Speaker 1
Play
8.213333
20
E mm' 'a mette dint' 'a casa, dicenno ca era l'infermiera... Pecché isso se credeva overo ca io stevo murenno.
30
030.wav
Speaker 1
Play
3.328
5
Madonna... quanto me faie schifo!
31
031.wav
Speaker 1
Play
4.458667
10
E se io stevo murenno overamente, tu chesto avisse fatto?
32
032.wav
Speaker 1
Play
3.818667
11
Ma pecché, tu murive e io nun avev' 'a magnà cchiu?
33
033.wav
Speaker 1
Play
2.794667
4
Nun m'avev' 'a sustené?
34
034.wav
Speaker 1
Play
2.453333
5
Ch'e rrose mmiez' 'a tavula?
35
035.wav
Speaker 1
Play
4.778667
8
Ma pecché, nun ero padrone d' 'e mmettere?
36
036.wav
Speaker 1
Play
2.133333
4
Quanto me faie ridere!
37
037.wav
Speaker 1
Play
6.4
22
Ma che me ne mporta 'e te, d' 'a figliola che t'ha fatto perdere 'a capa, 'e tutto chello ca me dice?
38
038.wav
Speaker 1
Play
3.818667
11
Ma tu te cride overo ca io ll'aggio fatto pe' te?
39
039.wav
Speaker 1
Play
3.498667
9
Ma io nun te curo, nun t'aggio maie curato.
40
040.wav
Speaker 1
Play
8.448
19
Na femmena comm' a mme, ll'he ditto tu e mm' 'o stai dicenno 'a vinticinc'anne, se fa 'e cunte.
41
041.wav
Speaker 1
Play
2.389333
5
Me sierve... Tu, me sierve!
42
042.wav
Speaker 1
Play
4.010667
7
E denare! E nun te l'avarria date?
43
043.wav
Speaker 1
Play
4.394667
6
Filume', tu afforza me vuo pògnere?
44
044.wav
Speaker 1
Play
3.242667
8
Ma pecché nun avev' 'a mangià, secondo voi?
45
045.wav
Speaker 1
Play
1.728
4
Qua sta 'a cena.
46
046.wav
Speaker 1
Play
4.245333
9
Quanno site venuto ogge p'urdinà 'a cena, ve ricurdate?
47
047.wav
Speaker 1
Poetry
3.712
5
Cient’anne arreto ch’era viva Vava,
48
048.wav
Speaker 1
Poetry
4.117333
5
nnante che ffosse Vartommeo Coglione,
49
049.wav
Speaker 1
Poetry
3.562667
6
dicea no cierto che l’auciello arava
50
050.wav
Speaker 1
Poetry
4.928
6
a ttiempo che sguigliaje lo Sciatamone.
51
051.wav
Speaker 1
Poetry
3.946667
6
Nc’era lo Rre Marruocco che s’armava,
52
052.wav
Speaker 1
Poetry
4.586667
5
panzera, lanza longa e toracone,
53
053.wav
Speaker 1
Poetry
3.946667
7
e po’ jeva a ttrovà li Mammalucche
54
054.wav
Speaker 1
Poetry
4.394667
6
co balestre, spigarde, e co ttrabucche.
55
055.wav
Speaker 1
Poetry
3.925333
6
Chillo fu tiempo che Berta filava,
56
056.wav
Speaker 1
Poetry
4.394667
6
co chillo doce vivere a l’antica!
57
057.wav
Speaker 1
Poetry
4.928
5
Portave brache, e nullo delleggiava!
58
058.wav
Speaker 1
Poetry
6.613333
12
Ogn’anno, il due novembre, c’è l’usanza per i defunti andare al Cimitero,
59
059.wav
Speaker 1
Poetry
3.605333
7
Si pe la via na femmena passava,
60
060.wav
Speaker 1
Poetry
4.202667
5
le dicevano: “Ddio la benedica!”.
61
061.wav
Speaker 1
Poetry
3.456
7
Mo, s’uno parla, e chella se corruzza.
62
062.wav
Speaker 1
Poetry
3.605333
7
Chi te pienze che ssia? Monna Maruzza.
63
063.wav
Speaker 1
Poetry
3.754667
6
O bell’ausanza, e ddove si’ squagliata?
64
064.wav
Speaker 1
Poetry
3.946667
7
Pecchè non tuorne, o doce tiempo antico?
65
065.wav
Speaker 1
Poetry
4.650667
7
Pigliave co lo bisco, a na chiammata,
66
066.wav
Speaker 1
Poetry
4.757333
6
cient’aucelluzze a no trunco de fico!
67
067.wav
Speaker 1
Poetry
4.501333
5
Le ffemmene, addorose de colata,
68
068.wav
Speaker 1
Poetry
6.421333
15
Ma chi te cride d’essere, nu ddio? Cca dinto, ‘o vvuò capì ca simmo eguae?
69
069.wav
Speaker 1
Poetry
3.733333
7
danzanno tutte ‘n chietta, (oh bona fede!)
70
070.wav
Speaker 1
Poetry
10.346667
12
Madonna, si ce penzo che paura! ma po’ facett’ un’anema ‘e curaggio
71
071.wav
Speaker 1
Poetry
3.712
6
Dove se trova mai tanta lianza!
72
072.wav
Speaker 1
Poetry
2.901333
7
Lo marito sì ccaro a la mogliera,
73
073.wav
Speaker 1
Poetry
3.946667
9
che a mano a mano ‘ntravano a na danza
74
074.wav
Speaker 1
Poetry
3.370667
4
co chella ciaramella tant’allera!
75
075.wav
Speaker 1
Poetry
3.626667
7
Vedive, a chioppa a chioppa, na paranza
76
076.wav
Speaker 1
Poetry
3.221333
5
co chell’antica e semprece manera!
77
077.wav
Speaker 1
Poetry
3.264
7
Lo viecchio a chillo tiempo era zitiello,
78
078.wav
Speaker 1
Poetry
3.968
7
co le brache stringate e ‘n jopponciello.
79
079.wav
Speaker 1
Poetry
2.389333
5
Chillo non era tiempo ammagagnato!
80
080.wav
Speaker 1
Poetry
4.821333
5
Le ffemmene assettate mmiezo chiazza,
81
081.wav
Speaker 1
Poetry
2.901333
5
non c’era n’ommo ch’avesse parlato,
82
082.wav
Speaker 1
Poetry
2.858667
6
ca vernava ‘n cajola la cajazza.
83
083.wav
Speaker 1
Poetry
2.624
6
era tenuto pe gallo de razza.
84
084.wav
Speaker 1
Poetry
2.88
6
Ll’uno co ll’autro lo mostrav’a dito:
85
085.wav
Speaker 1
Poetry
2.432
7
“Chillo che passa mo, chill’è lo zito!”
86
086.wav
Speaker 1
Poetry
3.264
5
Tutte le bon’ausanze so’ lassate!
87
087.wav
Speaker 1
Poetry
2.346667
5
Le rose mo deventano papagne!
88
088.wav
Speaker 1
Poetry
3.584
7
Lo vicenato, ‘n chietta e ‘n lebertate,
89
089.wav
Speaker 1
Poetry
2.730667
7
a chillo tiempo jevano a li vagne,
90
090.wav
Speaker 1
Poetry
7.829333
18
Pecché ‘ncopp’a sta terra femmene comme a te non ce hanna sta pé n’ommo onesto comme a me!…
91
091.wav
Speaker 1
Poetry
2.752
5
Si ddoce comme ‘o zucchero
92
092.wav
Speaker 1
Poetry
3.797333
7
E ghievano abbracciate a otto, a diece,
93
093.wav
Speaker 1
Poetry
5.397333
7
cchiù ghianche e rosse che le mmela-diece
94
094.wav
Speaker 1
Poetry
2.581333
6
Chella co la gonnella de scarlata
95
095.wav
Speaker 1
Poetry
3.370667
4
portava perne grosse comm’antrita.
96
096.wav
Speaker 1
Blog
9.386667
21
O Puorto 'e Napule è nu puorto mpizzato int''o Gurfo 'e Napule ca se spanne int''a custiera d''a cità 'e Napule.
97
097.wav
Speaker 1
Blog
3.413333
7
È uno d''e cchiù mpurtanti puorte d'Europa.​
98
098.wav
Speaker 1
Blog
9.194667
17
Melito 'e Napule (ditto Mêlito d''a ggente) è nu comune 'e 38.062 crestiane d''a pruvincia 'e Napule.
99
099.wav
Speaker 1
Blog
27.029333
50
O Regno 'e Napule (nomme ufficiale: Regno 'e Sicilia citeriore) è 'o nomme cu cui è canusciuto nu stato indipendente, ca esisteva tra 'o XIII e 'o XIX seculo e ca currispunneva a ll'attuale reggione 'e ll'Italia meridiunale, 'ncluse Abruzzo e parte d"o Lazzio, ma cca lassava fore 'a Sicilia
100
100.wav
Speaker 1
Blog
17.749333
28
O gurfo 'e Napule è na 'nzenatura d''o Mar Tirreno meriddionale, cumprèsce 'nfra 'a penisula flegrea a nord-ovest (capo Miseno) e 'a penisula surrientina a sud-est (punta Campanella)
101
101.wav
Speaker 1
Blog
9.642667
17
A Zona Nnustriale (o Gianturco) è nu rione 'e 6082 crestiani d''a part' 'e levante 'e Napule.
End of preview. Expand in Data Studio

Neapolitan Spoken Corpus (NSC)

A corpus of read Neapolitan speech for ASR evaluation, with a validated Neapolitan–Italian lexicon, LOSO fine-tuning splits, trained LoRA adapters, metric implementations, per-clip results, and error annotations.

This release supersedes the earlier 141-clip single-speaker version of this repository. The earlier release corresponds to Speaker S1 of the present corpus; the old audioData/ and transcripts.csv are replaced by data/audio/ and data/metadata.csv.

Corpus

  • 591 clips, 44.2 minutes total
  • 4 speakers (S1–S4): 141 / 150 / 150 / 150 clips
  • 3 text domains: Blog (147), Play (246), Poetry (198)
  • Audio: WAV, 16 kHz, mono, 16-bit PCM. Source recordings were captured at 44.1/48 kHz (WAV and M4A) and converted to 16 kHz WAV for this release.
  • Speakers: three male, one female; three adults aged 25–60 and one older adult (60+); all native Neapolitan speakers raised in Campania.
  • All texts are read speech from public Neapolitan-language sources (blog prose, theatrical plays, poetry).

Directory map

data/
  audio/                   591 WAV clips (16 kHz mono)
  metadata.csv             clip_id, filename, speaker, domain, duration_s,
                           n_tokens, reference_text
lexicon/
  lexicon_115.csv          validated 115-entry Neapolitan-Italian lexicon
  candidates_118.csv       pre-validation candidate list (118 entries)
  review_log.csv           native-speaker validation log (3 rejected,
                           6 amended, 1 variant-flagged)
  equivalence_classes.json orthographic normalization rules
splits/
  loso_fold{1,2,3,4}/      train.txt / val.txt / test.txt clip-ID lists
adapters/
  C{6,7,8,9}/fold{1,2,3,4}/  LoRA adapters (PEFT format), one per
                             condition x fold
metrics/
  ier.py, ier_alignment.py, normalize.py    IER implementation
  glotlid_ratio.py                          GlotLID-ratio metric
  scoring_pipeline.py                       WER/CER + end-to-end scoring
  reproduce.py                              reproduces headline numbers
results/
  per_clip_uncapped.csv    uncapped per-clip WER/CER, all conditions
  zero_shot/               per-clip results, conditions C1-C5
  lora/                    per-clip results, conditions C6-C9 (4 folds each)
annotations/
  error_analysis_pass1.csv 120-clip error annotations, first pass
  pass2/                   blind second-pass annotations, taxonomy card,
                           blind key, taxonomy-differences README
rebuttal/
  R1/                      zero-shot baselines for two non-Whisper systems
                           (SeamlessM4T-v2-large, MMS-1b-all)
  R3/                      lexicon-subsampling sensitivity analysis

Evaluation protocol (LOSO)

Fine-tuning conditions C6–C9 use leave-one-speaker-out cross-validation: each fold holds out one speaker's clips as the test set; the remaining three speakers' clips are split 90/10 into train/validation (seed 42). Fold definitions are in splits/. Reported corpus-level values pool the four held-out test sets, which together cover all 591 clips exactly once.

Reproducing the paper's numbers

cd metrics
python reproduce.py

This recomputes the headline corpus values (C1 WER 0.8124, C1 IER 0.1375, C9 IER 0.0612) from the raw per-clip references/hypotheses in results/ using only the released metric code and lexicon. See metrics/README.md for metric definitions and the full mapping to the results tables.

Python 3.10+ is required. requirements.txt pins the full environment used for the paper's experiments; reproduce.py itself needs only the standard library. The pinned torch==2.11.0+cu128 build installs from the PyTorch CUDA 12.8 index.

Adapters

Each adapters/C*/fold*/ directory is a PEFT LoRA adapter (adapter_config.json + adapter_model.safetensors; r=16, alpha=32, q_proj/v_proj). C6–C8 adapt openai/whisper-small; C9 adapts openai/whisper-medium. Load with:

from peft import PeftModel
from transformers import WhisperForConditionalGeneration

base = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small")
model = PeftModel.from_pretrained(base, "adapters/C6/fold1")

Dataset viewer

The viewer is configured to display data/metadata.csv (one row per clip). Audio files are under data/audio/ and are named by the filename column.

License and consent

This dataset is released under a Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) license. All participants provided informed consent, and no sensitive or personal information is included. See LICENSE.md.

Downloads last month
199