Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 423073449 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

tamil-turns

4,774 conversational turns from 115 real Tamil telephone calls — one speaker's continuous hold of the floor, with their own mid-turn pauses kept separate from the silence that ends the turn.

Companion dataset. santhosh-005/tamil-eot is the same source calls cut as 8 s windows with a binary complete / incomplete label — use that one to train an end-of-turn classifier. This dataset keeps whole turns with their silence structure — use it for turn-taking, mid-turn pauses, overlap and interruption, or timing analysis. Same corpus, same call-level split, different unit of analysis.

turns 4,774 — 16 kHz mono FLAC
source calls 115
audio 15.4 h
mid-turn pauses 9,709
train / dev / test 3,024 / 482 / 1,268
turn ends with a joined label 2,113
from datasets import load_dataset
ds = load_dataset("santhosh-005/tamil-turns")

datasets ≥ 4 needs torchcodec to decode the audio column. To avoid that dependency, read the FLAC bytes directly:

import io, soundfile as sf
from datasets import load_dataset, Audio

ds = load_dataset("santhosh-005/tamil-turns").cast_column("audio", Audio(decode=False))
x, sr = sf.read(io.BytesIO(ds["test"][0]["audio"]["bytes"]))

What it is good for

end-of-turn / endpointing the last silence_spans entry is the real end-of-turn
mid-turn pauses every earlier span is the speaker hesitating, not finishing
overlap and interruption speech_crosstalk measures the other leg talking across the turn; a negative eot_gap means they started before this speaker stopped
ASR / timing speech_s, word_rate, n_spans and the span list give per-turn timing

Structure

Split by call, not by turn — no call appears in two splits, and the split is the same one tamil-eot uses, so the two datasets agree on which calls are held out.

field meaning
id <call>_<L|R>_<turn_index>
audio 16 kHz mono FLAC — that speaker's leg only, never a mixdown
language ta
silence_spans {start, end} in clip time. The last is the end-of-turn; earlier ones are mid-turn pauses
messages the other speaker's preceding turn
duration speech portion, before the end-of-turn silence
eot_gap true length of the final silence
speech_s seconds actually spoken (duration minus the pauses)
word_rate n_words / speech_s
n_spans, n_absorbed VAD spans in the turn; backchannels folded in
pause_crosstalk, speech_crosstalk other-leg speech inside a pause / across the whole turn
transcript words apportioned to the turn — approximate, see Considerations
conversation_id, speaker_id, turn_index sort by turn_index to rebuild a call
split train / dev / test, by call
eot_bench_eligible final silence ≥ 0.2 s and no span > 5.0 s (4,322 rows)
label, source, sid joined from tamil-eot where a labelled boundary lands on this turn's end; null otherwise
split turns eligible
train 3,024 2,739
dev 482 437
test 1,268 1,146

silence_spans follows the layout used by livekit/eot-bench — last span is the end-of-turn, earlier ones are holds — so the rows are consumable by that kind of harness without conversion. eot_bench_eligible marks the rows meeting its usual release rules; the other 452 are fast handovers and long lapses, kept because interruption and timing work wants exactly those.

Dataset creation

Source

Derived from SPRING_INX Tamil R1 (SPRING Lab, IIT Madras, CC BY 4.0), which ships each call as one audio file per speaker, separately transcribed. That makes the speaker label structural rather than inferred — no diarization, and no speaker-error rate to propagate. It is also what makes turn construction possible at all: which file the audio came from is the speaker.

How turns were built

Turns come from floor occupancy over VAD spans confirmed against the human transcript, never from a turn classifier. Maximal same-side runs, with backchannels absorbed (< 3 words and < 1.0 s) and a split at any pause that is not a hesitation — over 2.0 s with the floor empty, or one the other speaker talks across.

Six gates then decide clean, and only clean turns are published: the turn opens and closes a transcript segment (so it never starts or ends mid-word), the other speaker is not talking across it, the handover is not overlapped, at least 1.0 s of speech, and at least 1.0 word per second of speech.

That gate is strict on purpose. Of 21,596 raw turns, 4,774 are clean — spot-listening the ungated set turned up fragments cut mid-word and "pauses" that were really the other speaker holding the floor.

Labels

This dataset is structural, not labelled — silence_spans, eot_gap and the floor layout are derived from audio and transcript, not from anyone's judgement, so no labelling was needed for them.

Where a labelled boundary from tamil-eot lands exactly on a turn's end, that label is joined in (2,113 of 4,774 turns). Those labels come from an audio LLM that agreed with a human listener on 97.5% of 197 blind-listened clips; the labeller bake-off and its limits are documented on the tamil-eot card. label is null on the remaining turns rather than guessed.

Considerations

  • One domain — narrowband two-party telephone conversations, task-oriented.
  • transcript is approximate. SPRING-INX ships segment-level transcripts, so words are apportioned across VAD speech seconds rather than force-aligned. No word-level timings ship, and transcript must not be treated as one.
  • Turns ending in overlap are excluded. The clean gate requires eot_gap >= 0, so turns the other speaker cut into are not here. They exist in the pipeline and can be rebuilt from it.
  • Trailing audio is capped at 5 s. For the 154 rows with a longer gap the final span is the capped length while eot_gap keeps the true value; all of those are eot_bench_eligible = False.
  • Turn length is long-tailed. Most turns are a few seconds, but genuine monologues run to several minutes — real, not a segmentation artefact.
  • Audio is conversational speech from a public corpus; no personally identifying metadata is included beyond what SPRING_INX ships.

Licensing and citation

The dataset — turn construction, the quality gates, and the splits — is by santhosh-005, released under CC BY 4.0.

@article{tamileot,
  author  = {Santhoshkumar V},
  title   = {TamilEOT: A Dataset and Model for Semantic End-of-Turn
             Detection in Tamil Telephone Speech},
  journal = {arXiv preprint arXiv:2609.05631},
  year    = {2026},
  url     = {https://arxiv.org/abs/2609.05631}
}

The source audio is from SPRING_INX Tamil R1 and remains the work of SPRING Lab, IIT Madras. It is redistributable here under CC BY 4.0, with attribution:

SPRING Lab, Indian Institute of Technology Madras. SPRING_INX Tamil R1. CC BY 4.0. https://asr.iitm.ac.in/SPRING_INX

Related

santhosh-005/tamil-eot the same calls as 8 s clips with a binary end-of-turn label — for training a classifier
santhosh-005/smart-turn-tamil the end-of-turn model trained on tamil-eot
smart-turn-livekit LiveKit Agents plugin for Smart Turn architecture models
github.com/santhosh-005/tamil-eot how all of it was built, including the experiments that failed
Downloads last month
167

Paper for santhosh-005/tamil-turns