# MInDS-14: multilingual spoken intent detection

## Facts

- Provider: Daniela Gerz, Pei-Hao Su, Razvan Kusztos, Avishek Mondal, Michal Lis, Eshan Singhal, Nikola Mrksic, Tsung-Hsien Wen, Ivan Vulic (PolyAI)
- License: CC-BY-4.0 (https://huggingface.co/datasets/PolyAI/minds14)
- Size: 18 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/PolyAI/minds14
- Access: public, readable without an account
- Category: Speech & Audio
- Source: https://arxiv.org/abs/2104.08524
- Page: https://scigantic.com/archive/e8f98b82-6aed-5aa6-8a44-c27d9141710f
- Keywords: Speech & Audio, Parquet, Hugging Face

## About

MInDS-14 is a training and evaluation resource for intent detection from spoken data. It covers 14 intents taken from a commercial system in the e-banking domain (such as card issues, balance and direct debit queries), with spoken examples in 14 language varieties: Czech, German, Australian, British and US English, Spanish, French, Italian, Korean, Dutch, Polish, Portuguese, Russian and Chinese. Each example pairs a recording with its transcription, an English translation and the intent label. It is widely used for spoken language understanding, multilingual and cross-lingual intent classification, and as a small audio-classification fine-tuning example. This repository has one folder per locale (for example en-US, fr-FR, zh-CN) plus an 'all' configuration, each holding Parquet shards with the audio, transcription, English transcription, intent class and language (measured 2026-11-14; the Hub reports about 3.9 GB). Load one locale with datasets.load_dataset('PolyAI/minds14', 'fr-FR'). PolyAI licenses the data under Creative Commons Attribution (CC BY 4.0); cite Gerz et al., 'Multilingual and Cross-Lingual Intent Detection from Spoken Data', 2021.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("PolyAI/minds14")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `all/train-00000-of-00002.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/e8f98b82-6aed-5aa6-8a44-c27d9141710f
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/e8f98b82-6aed-5aa6-8a44-c27d9141710f
- Over MCP: `get_schema_card` with id `e8f98b82-6aed-5aa6-8a44-c27d9141710f` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Speech & Audio datasets](https://scigantic.com/topics/speech-and-audio)
