1.

Why Professional
Speech

Most speech data fails in AI applications for reasons that converge on the same root causes:

Cannot reproduce the same conditions

Expression varies unpredictably across sessions

Emotion and prosody are incidental

Interpretation of instructions varies between speakers

Cannot re-record under changed conditions later

These are not failures of speaker ability — they are differences in professional capability.

Professional announcers, actors, and voice artists can:

Instantly translate instructions into vocal expression

Reproduce the same conditions repeatedly

Intentionally produce fine-grained variations

This capability makes speech treatable as a controllable variable.

2.

Speaker Network
& Selection

2.1 Speaker Attribute Design

For each project, we pre-specify:

Age Range: e.g., late 20s, early 50s

Gender & Voice Quality

Regional Background: Presence or absence of dialect influence

Speech Characteristics: Clarity, tempo, pitch range, prosodic range

This is not "casting" — it is data design.

2.2 Speaker Assignment Design

Cases requiring long-term speaker fixation

Cases requiring speaker rotation under aligned conditions

Cases varying only age or gender

Composition is structured according to comparison and learning objectives.

Principle

Control, Not Chance.

3.

Recording Design

The process of controlling speech as an engineering target.

3.1 Session Design

Each session explicitly designs:

Speaking Rate: Perceived tempo, not BPM

Pauses: Types and durations

Prosodic Range: Intonation variation

Emotional Arc: Start point, end point, and transition

Backchannels & Interruptions: Presence or absence

Critical: We do not specify emotions by "type."

Instead of "anger"

Calm → Discomfort → Frustration → Suppressed anger

Instead of "joy"

Surprise → Relief → Acceptance

Emotions are designed as continuous transitions, not categorical labels.

3.2 Instruction Method — Professional Premise

Professional speakers can receive mixed instructions in:

Emotional terminology

Performance terminology

Acoustic engineering terminology

This enables production of: same text, different non-verbal expressions, reproducible differences.

4.

Non-Verbal
Control in Speech

Non-verbal elements handled in professional speech data:

Breath & Breathing: Inhalation, breath pauses

Hesitation Sounds: Fillers, pauses, false starts

Backchannel Types & Timing

Sentence-Final Processing: Assertive vs. implied

Voice Onset & Decay

None of these are left to chance.

All are generated with: instruction, condition, and reproducibility.

5.

Annotation &
Structuring

5.1 Why We Avoid Simple Labels

Single-label emotion annotations like "anger" or "sadness" are not used.

Reasons:

Human emotions change over time

In speech, they overlap

When used for training, they cause overfitting

5.2 Structure We Adopt

Temporal Intervals: start / end

Continuous Quantities: Intensity, tension, etc.

Correspondence with Utterance Units

This enables learning not of "emotional states" but of expressive processes.

6.

Rights & Legal
Design

Professional speech data becomes entirely unusable in the future if rights design is weak.

M9 STUDIO requires:

Usage definition before recording

Explicit scope, region, and duration

Written documentation of reuse, derivatives, and retraining permissions

Linkage between consent content and audio samples

This creates a structure that withstands: EU research, international commercial use, and future model updates.

7.

Deliverables

Varies by project, but typically includes:

Audio Data: RAW / normalized

Non-Verbal Annotation

Metadata: Speaker attributes, conditions

Session Design Document

Annotation Guide

QC Report

Rights & Usage Summary

"Audio files only" is never delivered.

8.

Typical Use Cases

This business is chosen for the following applications:

High-Quality TTS
(especially emotion control)

Dialogue AI
(turn-taking, backchannels)

Avatars &
AI Characters

Robot
Speech I/O

Emotion Understanding &
Speech Behavior Analysis

The common thread: "We want to use the data we create for years."

9.

200+ Professional
Performers

Through partnerships with trusted Japanese production studios, we provide access to a diverse network of professional voice talent.

Performer Types

Voice Actors — Anime, game, and dubbing professionals

Actors — Stage and screen performers with vocal training

Announcers — Broadcast and corporate narration specialists

Narrators — Documentary, audiobook, and e-learning voices

Available Voice Categories

Narration & Announcement

Professional broadcast quality

Customer Service

Polite, clear guidance voices

Character & Animation

Expressive performance styles

AI Assistant

Neutral, natural conversational tones

Emotional Expression

Joy, anger, sadness with intensity control

Regional Dialects

Kansai, Tohoku, and other regional varieties

Demographics

Gender: Male, Female, Non-binary

Age Range: 20s through 70s

Languages: Japanese (native), with some bilingual performers

All performers are contracted through established production studios with proper rights management.

10.

Request Voice Samples

New sample library with 200+ performers launching soon.

Due to licensing and NDA requirements, voice samples are provided on request.

Can't wait? Contact us now for early access to voice samples matched to your requirements.

What We Need

Application type, voice style preferences, estimated data volume

What You Receive

Curated sample pack matched to your requirements

REQUEST EARLY ACCESS

Response within 2 business days. NDA available upon request.

11.

Why This Cannot Be Replaced

Controllable Speakers

Professional speakers treated as controllable variables

Re-Recording Capability

Can re-record under the same conditions any number of times

Continuous Emotion Design

Emotion designed as continuous structure, not categories

Non-Verbal Control

Non-verbal elements never left to chance

Long-Term Rights

Rights preserved for future use without degradation

M9 STUDIO's professional speech data business is not about collecting voices — it is about treating human expression as an engineering discipline.

This is not suited for requirements that are: cheap, fast, or high-volume.

However, for those who want to build speech AI that won't break in production — we can support from start to finish.

DISCUSS YOUR REQUIREMENTS

NEXT

Business 04

Demographically Structured Japanese Language Data

VIEW DETAILS