FLEURS-MN: Quality-Filtered Mongolian Speech and Text for TTS

2,009 n-way parallel sentences across 102 languages form the basis of this Mongolian speech dataset. It is a quality-filtered version of the FLEURS benchmark, cleaned via a 6-stage automated pipeline specifically for TTS training. The dataset was derived from google/fleurs by author btsee and last updated on 2026-05-23.

Use Cases

• Train Mongolian text-to-speech models based on the parallel speech and text data.
• Benchmark speech synthesis quality based on the quality-filtered audio samples.
• Fine-tune multilingual TTS systems based on the 102-language parallel sentence structure.
• Conduct linguistic analysis of Mongolian speech based on the transcribed sentences.

Strengths

• Derived from the established FLEURS benchmark, which contains 2,009 parallel sentences.
• Underwent a 6-stage automated quality filter calibrated for Mongolian TTS training.
• Specifically prepared for compatibility with modern TTS frameworks like oron-tts.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and dataset size are unknown, which may limit suitability assessment.
• License information is unknown, which may restrict commercial use.

Provenance

Source

Derived from google/fleurs config mn_mn.

Collection Method

Automated 6-stage quality filtering pipeline.

Freshness

Last updated 2026-05-23 12:06:36; freshness should be verified.