Datasetpublic

BrunoHays/Bangor-Miami-Spanish-English-Corpus

Bangor Miami Spanish-English Corpus The Bangor Miami Corpus is a naturalistic Spanish-English code-switching speech dataset collected by Jon Russell Herring at Bangor University. It captures spontaneous bilingual conversations recorded in Miami, Florida, involving proficient Spanish-English bilinguals across multiple speaker groups. Dataset description Total recordings 56 Total duration ~32 h Languages English (en), Spanish (es) Format MP3 audio… See the full description on the dataset page: https://huggingface.co/datasets/BrunoHays/Bangor-Miami-Spanish-English-Corpus.

sourceHugging Facecc-by-sa-3.0updated 4mo agoView on Hugging Face

0likes132downloads

Dataset Card

Bangor Miami Spanish-English Corpus

The Bangor Miami Corpus is a naturalistic Spanish-English code-switching speech dataset collected by Jon Russell Herring at Bangor University. It captures spontaneous bilingual conversations recorded in Miami, Florida, involving proficient Spanish-English bilinguals across multiple speaker groups.

Dataset description


Total recordings	56
Total duration	~32 h
Languages	English (`en`), Spanish (`es`)
Format	MP3 audio + plain-text transcript
Speaker groups	herring (16), maria (15), sastre (13), zeledon (12)

Each row is one full recording session and contains:

Column	Type	Description
`audio`	Audio	Raw waveform decoded at original sampling rate
`file_name`	string	Recording identifier (e.g. `herring1`, `sastre3`)
`transcript`	string	Full conversation transcript, one utterance per line, cleaned of CHAT annotation markers
`spanish_share`	float	Fraction of Spanish words in the recording (0 = all English, 1 = all Spanish), computed from word-level language-ID annotations in the corpus
`languages`	list[string]	Always `["en", "es"]`
`duration_s`	float	Recording duration in seconds

Spanish share distribution

spanish_share is computed as:

spanish_share = count(langid == "spa") / count(langid != "999")

where langid comes from the word-level TSV annotations shipped with the corpus and 999 marks punctuation tokens. The corpus spans from near-monolingual English (< 2 %) to near-monolingual Spanish (> 95 %), with a mean of ~34 % Spanish words.

Configurations

`default`

All 56 recordings (~32 h total).

`mixed`

A ~2.5 h subset of 5 recordings restricted to genuinely mixed conversations where 20%–80% of content words are Spanish. Recordings are drawn from all four speaker groups and cover the full 0.2–0.8 Spanish-share range.

Speaker	File	Duration	Spanish share
herring	herring17	~30 min	0.37
maria	maria20	~32 min	0.37
sastre	sastre8	~33 min	0.39
zeledon	zeledon4	~22 min	0.43
zeledon	zeledon14	~33 min	0.79

Source & citation

The original corpus was collected and transcribed at Bangor University. If you use this dataset, please cite the original work:

bibtex

@misc{bangor_miami,
  author    = {Deuchar, Margaret and Davies, Peredur and Herring, Jon Russell
               and Parafita Couto, Maria Carmen and Carter, Diana},
  title     = {Building bilingual corpora},
  booktitle = {Bilingualism: Basic principles and beyond},
  editor    = {Thomas, Enlli and Mennen, Ineke},
  year      = {2014},
  publisher = {Multilingual Matters},
  address   = {Bristol}
}

The corpus is available under CC BY-SA 3.0.