1.

Why Existing
Data Fails

Most acoustic and IR data fails to function adequately for AI applications for these reasons:

Measurement conditions don't match intended use

Spatial information is missing

Not designed for reuse

No alignment with simulation

Cannot be applied to physical robots

Especially in robotics and embodied AI, a frequent situation occurs:

"The measurement is correct, but it can't be used."

2.

Our Approach

We treat acoustics and IR not as byproducts of speech or acoustic engineering, but as primary input for spatial perception.

Therefore, the design target is not sound itself, but relationships.

Sound as a Perceptual System — not as audio engineering output.

Principle

Perception, Not Recording.

3.

Data Types

3.1 Impulse Response (IR)

Design targets:

  • Geometric relationship between source and receiver
  • Distance, height, angle
  • Reflective surface composition
  • Absorption and reflection characteristics

IR is not measured in isolation. We first define "what configuration, what reproduction target."

3.2 Environmental Acoustics

  • Indoor / outdoor
  • Quiet / noisy
  • Presence / movement of people
  • Overlapping machine sounds, ambient sounds

Environmental sound is not random noise. For AI, it directly connects to: direction information, distance information, priority judgment.

3.3 Spatial Audio

Explicitly designed:

  • Azimuth
  • Elevation
  • Movement trajectory
  • Occlusion and diffraction
4.

Acquisition Design

This is the core of the business.

4.1 Requirements Definition — What to Decide Before Acquisition

Target System

Robot hearing

XR

Spatial AI

Digital twin

Usage Phase

Simulation

Training

Evaluation

Deployment

Reproduction Target

Real space

Abstract space

Typical environment

Failure Modes to Avoid

Excessive reflection

Directionality misidentification

Distance estimation errors

4.2 Acquisition Geometry Design

Intentionally designed:

Source placement

Microphone placement

Movement patterns

Height differences

Fixed / variable conditions

This enables: same-space comparison, conditional learning, and regeneration.

5.

Measurement &
Reproducibility

5.1 Mandatory Recording Items

Acquisition date/time

Environmental conditions

Equipment used

Calibration conditions

Spatial dimensions

Position information

"Data that can't be recalled later" is not created.

5.2 Reproducibility Guarantee

Re-measurable under same conditions

Differential generation with changed conditions

Time-lapse changes also acquirable

This is the premise for research and long-term operation.

6.

Metadata Design

Acoustic and IR data loses meaning without metadata.

We design as standard:

Spatial dimensions

Source / receiver positions

Azimuth / angle

Distance

Environmental characteristics

Intended use

This enables cross-application use in: simulation, model training, and physical deployment.

7.

Robotics &
Embodied AI

This business has particular strength in the following domains:

Robot Sound Localization

Human Distance Estimation

Action Decision Making

Navigation Assistance

Environmental Change Adaptation

We create acoustic input that connects to "how to act" — not just "what is heard."

8.

Deliverables

Typically includes:

Raw Acoustic Data

IR Data

Spatial Metadata

Design Specification Document

Acquisition Conditions Document

Regeneration Premise Notes

We do not end with simple wav / IR file delivery.

9.

Why This Cannot Be Replaced

Use-Case Design

Can design for intended use before acquisition

Structural Treatment

Can treat space as structure

Reuse & Regeneration

Built with reuse and regeneration as premise

Robotics Understanding

Understands robotics implementation requirements

Cross-Modal Integration

Can integrate with non-verbal and linguistic data

The decisive difference: elevating acoustics and IR from "measurement" to "perception design."

M9 STUDIO's acoustic, IR, and spatial data business is not about collecting sound — it is about preventing AI from misunderstanding the world.

Not suited for: cheap one-time measurements or quick-turnaround data.

However, for those seriously building AI and robots that work in real environments — we can support from design to execution.

DISCUSS YOUR REQUIREMENTS

NEXT

Business 06

Multimodal Integrated Data Architecture

VIEW DETAILS