Skip to main content

About TOMCUBE

Turning multimodal data into verifiable scientific evidence

TOMCUBE works with universities, research institutes and enterprises on multimodal language data: study design, collection, processing, annotation and verification. Audio, video, text, eye tracking and physiological signals share one timeline, delivered as research-grade data that is reproducible, auditable and citable.

Founded
2019
Headquarters
Beijing, China
Partner institutions
10+
Languages
38

01 / Our story

It starts with one timeline

For seven years we have done one thing that keeps getting harder: keep signals from different sources trustworthy on a single timeline.

Founded
2019
Headquarters
Beijing, China
Offices
2
Legal entity
Tangchuan Technology (Beijing) Co., Ltd.

TOMCUBE was founded in 2019. At first we did one thing: put sound, picture and sensor signals on a stable, shared timeline. Seven years later the team has grown from a single recording room into a Beijing headquarters and a Shanghai joint laboratory, and the corpus has grown from the first hour of read speech to 12,400 hours of multimodal recordings.

What we do

We cover the full chain around a research question: study design, collection, processing and annotation, verification and delivery. The difference is not an equipment list but verification: synchronization error, annotator agreement and batch QA results are written into reports that can be re-checked and shipped together with the data.

How we work

Every project has research, engineering and ethics roles. Research translates the question into measurable variables; engineering makes collection and processing stable and reproducible; ethics owns informed consent, de-identification and the licensing chain. Deliverables are more than a set of files: they include collection logs, annotation manuals and verification conclusions.

How we prefer to be seen

Not as a data vendor, but as an external methods group for research teams. After delivery, the method should still be open to questioning, reproduction and citation. That is the only standard by which we judge whether a project is finished.

02 / Values

Four principles we do not trade away

They are not painted on a wall. They are written into every delivery report, every ethics review and every decision to re-collect a batch.

01

Reproducibility first

Data, collection logs, annotation manuals and analysis scripts are delivered together. Anyone, at any later date, can reproduce the same conclusions.

02

Verification before delivery

Synchronization error, annotator agreement and batch QA results all go into the delivery report. A failed batch is re-collected; we do not trade accuracy for schedule.

03

Ethics is part of the design

Consent, de-identification and the licensing chain are settled before collection starts, not added as a note after the data is already on a drive.

04

Alongside research teams

We design collection from the research question, not from an equipment list. We would rather be an external methods group than a data vendor.

03 / Capabilities and certifications

A complete chain, from collection to verification

We do not split the work across outsourced stages: study design, collection, processing, annotation and verification are done by one team, with a clear chain of accountability.

Capabilities

  • Lab-grade multi-stream synchronized capture
  • Field and real-world collection
  • Remote and crowdsourced collection
  • Multilingual participant recruitment
  • Transcription and forced alignment
  • Annotation scheme design and management
  • De-identification and privacy engineering
  • Verification reports and reproducibility audits

Certifications and compliance

  • ISO 27001 Information Security Management

  • ISO 20252 Research & Data Services

  • IRB / research ethics collaboration

Certificate copies are available on request. Projects involving personal data or minors require ethics review and a signed data processing agreement.

04 / Timeline

2019–2026: a timeline that keeps getting longer

From the first recording room to a Beijing headquarters and a Shanghai joint laboratory, the scale has changed. What has not is that every signal has to line up in time.

  1. 2019

    From one recording room

    TOMCUBE is founded in Beijing and builds its first synchronized audio-video capture chain for a phonetics project. Three people, with the equipment packed into a car trunk.

  2. 2020

    First multimodal corpus delivered

    The first multimodal corpus with speech, video and transcripts is completed. Synchronization error enters the delivery report and becomes standard practice for later projects.

  3. 2021

    Verification and QA system

    Double-blind annotation review and batch sampling are introduced. Annotator agreement and QA pass rate become hard gates before any delivery.

  4. 2022

    Shanghai Joint Laboratory opens

    The Shanghai joint laboratory begins operation, working with university teams on multimodal collection and verification methods and providing experimental space for language and cognition research in East China.

  5. 2023

    Open research program

    The first open study and annotation protocol are published, and internal measurement conventions start to become reproducible public documents.

  6. 2024

    Clinical and eye-tracking datasets

    The clinical conversation de-identification benchmark and the READ Chinese reading eye-tracking dataset are released, showing that privacy protection and research usability can both hold.

  7. 2025

    Classroom protocol and dialect benchmark

    The classroom multimodal annotation protocol reaches v2 and the dialect speech recognition robustness benchmark completes its first round, with partner institutions reaching 10+.

  8. 2026

    12,400 hours and MC-Sync

    Verified multimodal corpus reaches 12,400 hours. The MC-Sync synchronization benchmark is released, and collection and verification workflows become reproducible for external teams.

05 / Ecosystem

With labs, keeping the method in the field

From C9 university speech labs to automotive cabins, concert halls and language-neuroscience groups, we build collection, annotation and verification workflows together in real research settings.

C9 university speech labs

Speech capture · prosody & speech technology

Language university multimodal labs

Classroom & interview · multimodal annotation

Immersive audio perception labs

Spatial audio · perception experiments

Automotive intelligent multimodal

In-cabin multi-talker · noisy scenes

Music conservatory multimodal labs

Performance motion · audio & physiology

Language science & neuroscience

EEG / ERP · language processing

...and a growing set of language, cognition and human-computer interaction research settings

06 / Team

Four people, four kinds of scrutiny

Research, annotation, engineering and ethics each have an owner. Every delivery passes through all four perspectives.

L

Lin Che

Founder · Chief Scientist

Phonetics and multimodal corpora; leads study design, synchronization verification and the open research program.

Z

Zhou Yuan

Research Director

Multimodal annotation and agreement assessment; owns annotation schemes, open benchmarks and joint research.

C

Chen Yan

Head of Data Engineering

Collection pipelines, multi-stream synchronization and processing tools; accountable for lab and field execution quality.

S

Su Yao

Head of Ethics & Compliance

Informed consent, de-identification and licensing chains; leads ethics review collaboration and data compliance.

07 / Offices

Working from Beijing and Shanghai

The Beijing headquarters leads study design, data engineering and verification delivery. The Shanghai joint laboratory handles multimodal collection methods and equipment validation. Visits are by appointment.

Beijing

Headquarters

Room 247, 2F, East Building, No. 015 Shui Niufang, Dougezhuang Township, Chaoyang District

Shanghai

Joint laboratory

Shanghai · Joint Laboratory (by appointment)

Work with us

Bring us your next research question

Whether you are preparing a grant application, building a data asset or reviewing an existing corpus, we can start with the question itself.

Reply within one business day · NDA and data licensing supported · Dedicated academic track