Datasets:
File size: 1,813 Bytes
7844133 f406d40 7844133 f406d40 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 | ---
license: apache-2.0
language:
- en
task_categories:
- audio-to-audio
tags:
- full-duplex
- spoken-dialogue
- turn-taking
- synthetic
---
# FD_data_v1 — synthetic full-duplex spoken conversations
A small demo batch of fully synthetic two-speaker conversations with
scripted turn-taking events and construction-time ground-truth timestamps.
Voices are synthesized with [Kokoro-82M](https://huggingface.co/hexgrad/Kokoro-82M)
(Apache-2.0); transcripts are synthetic and contain no personal data.
Each row is one two-speaker conversation:
- `audio`: stereo 24 kHz WAV — **channel 0 = user, channel 1 = assistant**.
Overlap (backchannels, interruptions) is real simultaneous audio across
channels.
- `policy` / `instruction`: turn-management conditioning
(none | interrupt | listen | backchannel | continue | acknowledge —
taxonomy after Instruct-FD, arXiv 2607.20460).
- `turns`: JSON — per turn: speaker, text, absolute start/end, word-level
onset/end timestamps (MMS forced alignment), `cut` + `unspoken` tail when
the turn was interrupted (the full reference text is retained).
- `events`: JSON — scripted-event ground truth with realized timestamps
(pause / assistant_backchannel / assistant_interrupt / user_backchannel /
user_interrupt). Timestamps are constructed, not estimated.
- `fidelity`: JSON — per-clip whisper-large-v3-turbo script-fidelity ratios.
- `verify_pass` / `verify_checks`: deterministic verdict — the event
structure is re-derived from the assembled audio via Silero VAD and checked
against the script (bc timing/shortness, interrupt onset+cut, continue
no-yield, acknowledge yield+adapted, pause no-takeover, listen
no-talk-over).
Timing bands (gaps, reaction latencies) are v1 PLACEHOLDERS, not yet
calibrated against real-corpus statistics.
|