Datasets:

livekit
/

eot-bench-data

license: cc-by-4.0
configs:
  - config_name: all
    default: true
    data_files:
      - split: validation
        path: data/*/validation-*.parquet
  - config_name: en
    data_files:
      - split: validation
        path: data/en/validation-*.parquet
  - config_name: ar
    data_files:
      - split: validation
        path: data/ar/validation-*.parquet
  - config_name: de
    data_files:
      - split: validation
        path: data/de/validation-*.parquet
  - config_name: es
    data_files:
      - split: validation
        path: data/es/validation-*.parquet
  - config_name: fr
    data_files:
      - split: validation
        path: data/fr/validation-*.parquet
  - config_name: hi
    data_files:
      - split: validation
        path: data/hi/validation-*.parquet
  - config_name: id
    data_files:
      - split: validation
        path: data/id/validation-*.parquet
  - config_name: it
    data_files:
      - split: validation
        path: data/it/validation-*.parquet
  - config_name: ja
    data_files:
      - split: validation
        path: data/ja/validation-*.parquet
  - config_name: ko
    data_files:
      - split: validation
        path: data/ko/validation-*.parquet
  - config_name: nl
    data_files:
      - split: validation
        path: data/nl/validation-*.parquet
  - config_name: pt
    data_files:
      - split: validation
        path: data/pt/validation-*.parquet
  - config_name: tr
    data_files:
      - split: validation
        path: data/tr/validation-*.parquet
  - config_name: zh
    data_files:
      - split: validation
        path: data/zh/validation-*.parquet

LiveKit EoT Benchmark

This dataset contains a minimal public subset of the LiveKit end-of-turn benchmark. It is sampled from the full release-ready dataset preserved on the full revision.

Links

Blog post: Solving end-of-turn detection
Benchmark code: github.com/livekit/eot-bench

The sample is deterministic: up to 400 turns per language were sampled without replacement using seed 20260603. Languages with fewer than the requested sample size include all available rows: ar (381 available), id (396 available), ja (356 available), ko (377 available), zh (344 available).

Configs

all: all available sampled languages.
Language configs such as en, de, and ja: one language only.

All configs use the same schema. English rows include language = "en".

Schema

id: globally unique turn id.
audio: 16 kHz mono WAV audio.
language: normalized language code.
duration: audio duration in seconds.
silence_spans: candidate silence spans with start and end.
words: word-level transcript timings.
messages: prior conversation messages.

Rows whose final end-of-turn silence span is shorter than 0.2s are excluded from this release split, so every turn can be evaluated at a 0.2s score point. Rows with any silence span longer than 5.0s are also excluded so the benchmark focuses on ordinary endpointing pauses.

Languages

Language	Rows
`en`	400
`ar`	381
`de`	400
`es`	400
`fr`	400
`hi`	400
`id`	396
`it`	400
`ja`	356
`ko`	377
`nl`	400
`pt`	400
`tr`	400
`zh`	344
`all`	5454