Spanish–English code-switching speech data
Spanglish as it is really spoken: US Hispanic and Mexican speakers moving between Spanish and English inside one sentence, “vamos a la meeting y luego I’ll call you”. Every word is transcribed as said and tagged with its language, and every switch point is marked, so recognition does not break at the language boundary.
Specification
- Speakers
- Bilingual US Hispanic speakers and Mexican speakers who use English every day, balanced across gender and age groups
- Settings
- Conversations with family and friends, work talk, customer-service calls, voice notes and interviews
- Audio
- 8 kHz telephone or 16 kHz wideband, 16-bit WAV
- Annotation
- Verbatim transcript, a language tag on every word, switch points, timestamps and speaker turns
- Spelling
- Spanish and English words in their standard spelling, accents and ñ kept
- Quality check
- A second bilingual reviewer checks every transcript and every language tag
Example of the data
What you see, and the file you get.
- Said
- Vamos a la meeting y luego I’ll call you.
- Language per word
- es es es en es es en en en
- Switches
- 3 (es→en, en→es, es→en)
| audio_url | word_index | word | language | start | end | switch |
|---|---|---|---|---|---|---|
| …/audio/cs_0001.wav | 1 | Vamos | es | 0 | 0.34 | |
| …/audio/cs_0001.wav | 2 | a | es | 0.34 | 0.41 | |
| …/audio/cs_0001.wav | 3 | la | es | 0.41 | 0.52 | |
| …/audio/cs_0001.wav | 4 | meeting | en | 0.52 | 0.96 | es→en |
| …/audio/cs_0001.wav | 5 | y | es | 0.96 | 1.05 | en→es |
| …/audio/cs_0001.wav | 6 | luego | es | 1.05 | 1.38 |
Sample rows. Also as JSON or Excel; audio as WAV or MP3.
Contact us to get the dataset
Tell us what you need. We walk you through the data, its formats and how it is checked on a short call.
What it is used for
Speech recognition that survives switching
Train ASR that keeps both languages right inside one sentence.
Bilingual voice agents
Serve customers who start in Spanish and finish in English, without asking them to pick one.
Call and meeting transcription
Transcribe mixed calls and meetings without losing either language.
Benchmarks
Measure how well your model handles switching, word by word.
Questions
How is mixed speech written down?
Exactly as said: each word in its standard Spanish or English spelling, tagged with its language, with switch points marked.
Which Spanish is spoken?
Mostly Mexican and US Hispanic Spanish, including speakers from other Latin American backgrounds living in the US. Tell us the mix your users speak.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples