PolySpeech-100-v1: Multilingual Speech Dataset Covering 110 Languages

A large-scale multilingual speech dataset encompassing 110 languages, created by author 'youngseng' and last updated on May 23, 2026. The dataset is structured as 110 separate Parquet files, each corresponding to a specific language subset. A standalone Python script is provided to reconstruct the original dataset hierarchy from these files.

Use Cases

• Training multilingual automatic speech recognition (ASR) models based on the 110-language coverage.
• Developing text-to-speech (TTS) systems for low-resource languages based on the per-language data subsets.
• Benchmarking speech model performance across diverse languages based on the dataset's multilingual structure.
• Conducting linguistic or acoustic analysis of speech patterns across different language families.

Strengths

• Covers 110 languages, providing a wide multilingual scope.
• Structured into separate Parquet files per language, which may simplify language-specific workflows.
• Includes a provided Python script for reconstructing the dataset hierarchy.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and total size are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Freshness

Last updated 2026-05-23 07:38:43; freshness should be verified.