OpenSTBench Paired Set: LibriTTS-Based Speaker Preservation Evaluation Data

A paired speaker set constructed from the LibriTTS corpus for evaluating speaker preservation in speech-to-speech translation systems. The dataset supports the OpenSTBench multidimensional benchmark, as detailed in a 2025 paper. It was created by author ayj111 and last updated on June 4, 2026.

Use Cases

• Benchmarking speaker preservation performance in speech-to-speech translation models based on paired audio samples.
• Training or fine-tuning models to retain speaker identity during translation, based on the paired speaker set structure.
• Conducting ablation studies on which system components affect speaker characteristic transfer.

Strengths

• Specifically designed for a multidimensional benchmark (OpenSTBench) for speech translation.
• Built upon the established LibriTTS corpus, suggesting a foundation in a known audio dataset.
• Last updated on June 4, 2026, indicating recent maintenance.

Limitations

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

Provenance

Source

Derived from the LibriTTS corpus.

Collection Method

Constructed for the OpenSTBench evaluation benchmark.

Freshness

Last updated 2026-06-04 08:41:25.