About AI-SQ Engineering

Turning Sound Quality Evaluation into Engineering for Product Development

AI-SQ Engineering is an engineering service that combines psychoacoustics, acoustic analysis, subjective

listening evaluation, and AI technologies to support the evaluation, improvement, and design of product

sounds. Conventional product sound assessment has often focused on physical measurements such as sound

pressure level and frequency analysis.

However, sounds with the same sound pressure level can create very different impressions. Established

psychoacoustic metrics may also be unable to describe fully the product-specific characteristics of a sound

or the impression created by sounds that change over time.

AI-SQ Engineering connects acoustic data with human evaluation to identify the sound characteristics that

matter for each product.

What Is AI-SQ?

The “SQ” in AI-SQ stands for Sound Quality.

AI-SQ Engineering is not simply a system for automating sound quality evaluation. It is an integrated

approach that treats sound analysis, the relationship between sound and human perception, evaluation

model development, exploration of improvement directions, and sound design as one continuous process.

Sound Analysis

We analyse product sounds and extract the frequency components, temporal evolution, fluctuations,

periodicity, and other characteristics that define them.

Sound Quality Evaluation

We combine psychoacoustic metrics such as loudness, sharpness, roughness, fluctuation strength, and

tonality with our own metrics, classification methods, and product-specific evaluation criteria.

AI Modeling

We train AI models using analytical results and subjective listening data to classify sounds, predict

impressions, and identify the most influential features.

Sound Design

Based on the evaluation results, we examine which components should be reduced, retained, or emphasised, and how the temporal development or timbre should be shaped.

Key Characteristics of AI-SQ Engineering

1. We Build an Evaluation Method for Each Product

It is difficult to evaluate every product sound using a single universal metric.

The characteristics that attract attention differ among vehicle acceleration sounds, motor sounds, air-

conditioning sounds, vacuum cleaner sounds, alarming sounds, and many other types of product sound.

AI-SQ Engineering selects the analytical methods and evaluation criteria according to the target product,

usage conditions, and purpose of the assessment.

2. We Evaluate Both Frequency Structure and Temporal Structure

The impression of a sound is not determined by its frequency spectrum alone.

The way a sound begins, develops, fluctuates, and decays, as well as how it changes with operating

conditions, can influence impressions such as responsiveness, comfort, perceived quality, or power.

For acceleration sounds and rotating machinery sounds, the movement of order components with

changing rotational speed and the temporal development of timbre are also important.

AI-SQ Engineering integrates waveform, spectrum, and time-frequency characteristics in the evaluation.

3. We Connect Human Evaluation with Analytical Results

A change in an analytical metric alone does not demonstrate that product value has improved.

What matters is how that change relates to human impressions and preferences. We link the results of

subjective listening and sensory evaluations with acoustic features to identify the factors that shape the

impression of a product sound.

4. We Use AI to Support Design, Not to Replace Human Judgement

AI is well suited to finding relationships within large sets of acoustic features and evaluation data that may

be difficult for people to detect directly.

Final judgements about whether a sound is desirable or appropriate for a brand still require human

expertise. AI-SQ Engineering therefore uses AI as a tool for analysis, comparison, prediction, and design

support rather than as a substitute for human decision-making.

The AI-SQ Engineering Process

STEP 1 – Define the Objective

We clarify the target product, usage situation, current issues, and the intended sound impression. In

addition to requirements such as making a sound quieter, we define sensory goals such as creating a more

premium or powerful impression or reducing unpleasantness.

STEP 2 – Acoustic Measurement and Data Review

We review recordings, acoustic measurement data, operating conditions, and product specifications.

Existing measurement data may be used, or new recording and evaluation conditions may be designed.

STEP 3 – Acoustic and Psychoacoustic Analysis

We calculate physical acoustic quantities, waveform characteristics, frequency characteristics, time-

frequency characteristics, and psychoacoustic metrics. Depending on the target sound, we also analyse

variability, periodicity, temporal envelopes, and order structures.

STEP 4 – Subjective Listening Evaluation

When required, we conduct listening evaluations using methods such as paired comparison, magnitude

estimation (ME), and semantic differential (SD). We define evaluation terms appropriate to the product

sound and investigate the relationship between acoustic features and human impressions.

STEP 5 – Build AI and Statistical Models

Using analytical results and subjective listening data, we classify sounds, predict evaluations, and identify

influential features. The objective is not to use a complex model for its own sake, but to produce results

that can be translated into design decisions.

STEP 6 – Propose Directions for Improvement

Based on the evaluation results, we identify components to reduce or emphasise and propose how the

temporal development should be adjusted. When useful, we create comparison stimuli through sound

processing or sound design.

STEP 7 – Validation

We re-evaluate improved sounds or prototypes to confirm whether the intended sound quality has been

achieved.

Examples of Sounds We Address

AI-SQ Engineering is not limited to a particular product field. We work with a wide range of product sounds

  • through which people perceive the condition, quality, function, and character of a product.
  • Interior and exterior vehicle sounds
  • Operating sounds of engines, motors, fans, and other machinery
  • Operating sounds of household appliances and building equipment
  • Sounds associated with food and beverages
  • Notification sounds from health devices and medical-related equipment
  • Operating sounds from switches and mechanical components
  • Warning, notification, and interface sounds
  • Time-varying product sounds and sounds composed of multiple elements

Even within the same sound category, the required sound quality differs according to the product purpose,

usage environment, users, and brand. Rather than applying one predetermined method, AI-SQ Engineering

configures the evaluation metrics, listening evaluation methods, and analytical conditions for each target

product.

What AI-SQ Engineering Aims to Achieve

AI-SQ Engineering is not a service that merely measures sound and reports evaluation results.

Our goal is to transform the knowledge obtained from acoustic analysis into a form that can be used by

product engineers, CAE specialists, designers, and development managers.

In the future, the following functions will be integrated step by step:

  • Automatic extraction of acoustic features from WAV data
  • Sound quality evaluation using psychoacoustic metrics
  • AI-based sound classification and evaluation prediction
  • Visualisation of product sounds
  • Proposal of improvement conditions
  • Generation of product sounds using generative AI
  • Automatic evaluation and comparison of generated sounds
  • Support for final human sound quality judgements

By creating a continuous cycle of analysis, evaluation, design, and generation, we propose a new process for product sound development.

Consultation and Project Enquiries

Product sound challenges differ according to the development stage and the product itself.

Projects can be configured according to their purpose, including analysis only, subjective listening

evaluation only, re-evaluation of existing data, development of evaluation metrics, examination of AI

models, and sound design. We observe strict confidentiality and can support product development projects that cannot be disclosed publicly.