WikIPA

Dataset Description

WikIPA is a multilingual benchmark dataset designed for speech-to-IPA (STIPA) transcription, linking spoken audio with International Phonetic Alphabet (IPA) transcriptions.

The dataset integrates two large-scale community-driven resources:

• WikiPron — human-curated IPA pronunciations extracted from Wiktionary
• Lingua Libre — crowdsourced recordings of spoken lexical items

By connecting these two resources, WikIPA provides a dataset that links speech audio to phonetic representations, enabling evaluation of models that transcribe speech directly into IPA.

The dataset supports both:

• Broad (phonemic) IPA transcriptions
• Narrow (phonetic) IPA transcriptions

WikIPA is introduced in the paper:

WikIPA: Integrating WikiPron and Lingua Libre for Multilingual IPA Transcription

The dataset contains 289,694 audio–IPA pairs across 78 languages, making it one of the largest multilingual resources for speech-to-IPA evaluation.

Citation

If you use WikIPA or this repository, please cite:

@inproceedings{cassotti-etal-2026-wikipa,
  title = {WikIPA: Integrating WikiPron and Lingua Libre for Multilingual IPA Transcription},
  author = {Cassotti, Pierluigi and Suchardt, Jacob Lee and Cristofaro, Domenico De},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026)},
  month = {May},
  year = {2026},
  pages = {8465--8475},
  address = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy and Krek, Simon and Toral, Antonio},
  doi = {10.63317/2am4iw3bfhjb},
  abstract = {We present WikIPA, a new multilingual benchmark designed for automatic speech-to-IPA (STIPA) transcription. By integrating human-curated IPA transcriptions from WikiPron with spoken recordings and metadata from Lingua Libre, WikIPA connects textual phonetic representations with real speech across 78 languages. This open resource supports both broad (phonemic) and narrow (phonetic) transcription tasks, enabling fine-grained evaluation of multilingual phonetic transcription systems. WikIPA provides over 289,000 paired entries and serves as a large-scale foundation for STIPA. We benchmark several state-of-the-art STIPA systems, including MultIPA, (Lo)WhIPA, and ZIPA. Results show that ZIPA achieves the lowest mean error rates across most languages, outperforming Whisper- and Wav2Vec-based baselines. Error analyses reveal that remaining discrepancies largely stem from minor phonetic confusions rather than complete transcription failures, emphasizing the challenge of modeling fine-grained articulatory variation. WikIPA thus establishes the first systematic, multilingual evaluation framework for speech-to-IPA transcription and highlights the potential of combining open, community-driven resources to advance STIPA evaluation.}
}

---

Dataset Summary

Property	Value
Languages	78
Total samples	289,694
Audio duration (avg)	1.15 seconds
Speakers	962
Task	Speech → IPA transcription

Each entry corresponds to a single spoken lexical item recorded in Lingua Libre and linked to IPA pronunciations from WikiPron.

---

Dataset Structure

The dataset consists of audio recordings paired with IPA transcriptions and metadata.

Typical fields include:

• audio – speech recording
• lexical_item – word or phrase spoken
• ipa_broad – phonemic transcription (when available)
• ipa_narrow – phonetic transcription (when available)
• language – language code
• speaker – identifier of the speaker
• dialect – dialect information when available

Example entry:

{
  "audio": "...",
  "lexical_item": "domingo",
  "ipa_broad": "[d o m i N g o]",
  "ipa_narrow": "[d̪õmiŋgo]",
  "language": "spa",
  "speaker": "Eavqwiki"
}

A lexical item may have multiple possible IPA transcriptions, reflecting pronunciation variants or dialectal differences.

---

Data Splits

The dataset is divided into training and test splits.

Split	Examples
train	231,755
test	57,939
total	289,694

The test split is stratified by language to ensure balanced evaluation across languages.

---

Languages

WikIPA covers 78 languages from multiple language families.

The dataset includes languages with varying phonological complexity and different levels of phonetic annotation detail.

Both broad and narrow IPA transcriptions are available depending on the language and Wiktionary annotations.

---

Usage

Load the dataset using datasets:

from datasets import load_dataset

dataset = load_dataset("pierluigic/WikIPA")

Example:

sample = dataset["train"][0]

print(sample["audio"])
print(sample["ipa_broad"])
print(sample["language"])

---

Tasks

The dataset is intended for:

• Speech-to-IPA transcription (STIPA)
• Universal phone recognition
• Phonetic modeling
• Cross-lingual speech modeling
• Evaluation of multilingual phonetic transcription systems

---