Sample data
Look at the data before we talk
The sample pack comes from a real production batch: 40 episodes, 8 synchronized streams and 4 collection environments. Open data formats, annotation schema, QC report and reading scripts are available for evaluation without an NDA.
00:12.480 Okay, let us put this chart on the group table, and then—
aligned1.4 ms
Cross-stream sync
κ 0.91
Agreement
98.7%
Batch pass
- Episodes
- 40
- Synchronized streams
- 8
- Collection environments
- 4
- Sample pack
- 2.4 GB
01 / Inside the pack
What one episode contains
Each episode is a complete multimodal event: raw data streams, alignment metadata, annotation layers and QC records.
| Stream | Specification | Notes |
|---|---|---|
| Audio | 48 kHz / 24-bit · 4 channels | Close-talk + far-field microphone array, hardware-synchronized |
| Video | 1920 × 1080 · 60 fps | Front + side cameras with timecode overlay |
| Depth | 848 × 480 · 30 fps | Active stereo, 0.3–4 m range |
| Eye tracking | 120 Hz · screen/scene | Fixation points, pupil diameter, blink events |
| EEG | 1,000 Hz · 32 channels | Aligned to language stimulus event markers |
| Motion | 120 Hz · 21 keypoints | Hand and upper-body posture |
| Transcript | Word-level · timestamps | Human-corrected, precisely aligned to audio |
| Annotation | Event segments + labels | ELAN / JSON dual format |
manifest.json · excerpt
{
"episode_id": "EP_0042_CLASSROOM",
"environment": "primary_school",
"duration_s": 612.4,
"streams": [
{ "name": "audio", "rate": 48000, "channels": 4 },
{ "name": "video", "fps": 60, "resolution": "1920x1080" },
{ "name": "gaze", "hz": 120 },
{ "name": "eeg", "hz": 1000, "channels": 32 }
],
"sync": { "reference": "audio", "max_offset_ms": 2.1 },
"consent": { "irb": "TC-2026-014", "scope": "research" }
}File paths, checksums, device parameters and license scope are all written into the manifest, so analysis scripts can read them directly.
02 / Annotation
Annotation schema and reliability
Annotation is not an afterthought. The schema is fixed before collection, annotators are trained and assessed, and double-blind review produces agreement metrics you can cite in a paper.
| Layer | Granularity | Example labels | Agreement |
|---|---|---|---|
| Transcript | Word | Pinyin, characters, pauses, repairs | κ ≥ 0.92 |
| Discourse act | Turn | Question, response, instruction, restatement | κ ≥ 0.88 |
| Paralanguage | Segment | Intonation, stress, speech rate, emotion | κ ≥ 0.84 |
| Gaze | Fixation | AOI, first fixation duration, regression | Automatic + manual spot-check |
| Events | Event segment | Start, end, participant, object | κ ≥ 0.90 |
03 / Files
Delivery checklist
The sample pack uses the same directory structure and metadata conventions as a full delivery, so you can move to production data without friction once evaluation is done.
| File | Contents | Size |
|---|---|---|
| episodes/ | 40 episodes with 8 raw streams and alignment metadata | 2.1 GB |
| annotations.json | Layered annotation and timestamps, with annotator ID and version | 1.9 MB |
| manifest.json | Relative paths, checksums and device parameters for every file | 24 KB |
| qc-report.pdf | Sync error distribution, agreement statistics and spot-check results | 1.2 MB |
| annotation-spec.pdf | Annotation manual v2.4, with label definitions and boundary examples | 860 KB |
| README.md | Directory structure, reading examples and citation format | 12 KB |
How to get it
Four steps to the sample data
- 01
Submit a data request
Describe your research question, use case and data scope; we reply within one business day.
- 02
Confirm the license
Research license / commercial license / evaluation license, followed by the data use agreement.
- 03
Receive a download link
Delivered through an encrypted link with checksums and reading scripts.
- 04
Get support
Optional data walkthrough and reading support to help with your first analysis.
Reply within one business day · Sample pack is free
Start a project
Need data tailored to your research question?
Tell us the languages, scenarios, devices and annotation requirements. We will come back within one business day with a collection and validation plan, plus a comparable project delivery sample.
Reply within one business day · NDA and data licensing supported · Dedicated academic track