Eot Bench Data: LiveKit End-of-Turn Detection Benchmark Sample

A 2026-06-09 updated benchmark sample for end-of-turn detection in speech. The dataset is a deterministic sample of up to 400 turns per language from the full LiveKit benchmark, with languages having fewer samples including all available data. It was created by LiveKit to support the development and evaluation of models that detect when a speaker has finished a turn in conversation.

Use Cases

• Training end-of-turn detection models based on the described speech turn data.
• Benchmarking model performance on a multilingual task as suggested by the description of multiple languages.
• Analyzing speech patterns related to conversational turn-taking using the sampled audio data.

Strengths

• Deterministic sampling with a fixed seed (20260603) ensures reproducibility.
• Includes data from multiple languages, as indicated by the description of sampling per language.
• Sourced from a full, release-ready dataset preserved by the author, LiveKit.

Limitations

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

Provenance

Source

livekit

Collection Method

Sampled without replacement from the full LiveKit end-of-turn benchmark dataset.

Freshness

Last updated 2026-06-09 11:04:57; freshness should be verified.