SpeechEditBench: A Bilingual Multi-Attribute Speech Editing Benchmark

SpeechEditBench is a bilingual benchmark for instruction-guided speech editing, created by DiscreteSpeech. Each example provides source speech and a natural-language editing instruction to evaluate if a system can apply the requested edit while preserving lexical content. The dataset page was last updated on 2026-06-05.

Use Cases

• Benchmarking speech editing models based on natural-language instructions.
• Training instruction-following AI systems for audio modification.
• Evaluating the preservation of lexical content during speech edits.
• Researching cross-lingual (bilingual) capabilities in speech processing models.

Strengths

• Designed as a benchmark with a specific evaluator, suggesting a structured testing framework.
• Bilingual and multi-attribute nature indicates coverage of multiple languages and editing aspects.
• Includes natural-language instructions, providing a direct interface for human-guided editing tasks.

Limitations

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

Provenance

Source

DiscreteSpeech, hosted on Hugging Face.

Collection Method

Likely created for research purposes, as indicated by the associated paper and code repository.

Freshness

Last updated 2026-06-05 00:20:09; freshness should be verified.