リサーチ
Eleven v3（アルファ版）、ついに登場 — 感情まで伝える革新的TTSモデル
Eleven v3は、これまでで最も表現力豊かなテキスト読み上げモデルです
Introducing Voice Design v3
Discover Voice Design v3: create unique AI voices with ease. Describe your desired voice, get three instant options, and deploy for creators, businesses, and developers.
We're excited to announce Voice Design v3. The new release makes creating voices quicker, easier, and more intuitive than ever before.
Voice Design v3 is ideal for creators, businesses, and developers who need specific voices for their project. All you have to do is describe your voice, instantly get three options, choose your favorite, and start using it right away.
Give it a try and see how easy voice creation can be.
Why Voice Design Matters
When we launched Voice Design, the goal was simple: let any creator generate a bespoke voice — no studio booking, no audio library deep dives. Version 3 takes that further, offering more control, faster iteration, and a smoother path from concept to final audio.
Projects often require dozens of voices: a calm narrator, an anxious side character, a few game NPCs, maybe even a talking raccoon. Searching for “something close enough” slows teams down.
Voice Design v3 removes that friction. Type a description, generate three candidates, choose one, and move on — all while paying only for the characters in your prompt, not per sample.
What’s New in v3
- Precision audio quality: Add terms like “perfect audio quality” (or any other grade) to your prompt — v3 will honor them. Want a tin-can radio effect instead? Just say so. You control the polish.
- Smarter prompting engine: v3 handles nuanced cues like “middle-aged New Yorker with rising intonation and a half-smile” without introducing artifacts or losing clarity. Behind the scenes, we’ve expanded the latent space and retrained on broader accent data, making rare combinations sound natural.
Two design modes
- Realistic Voice Design for lifelike performances. Example: “An Indian woman with a soft, calm delivery.”
- Character Voice Design for anything from elves to elder gods.
One click, three voices
Hit Generate and v3 returns three distinct voices. Keep the one you like—it fills a voice slot—and discard the rest. No queue. No extra cost.
Design a voice in three steps
- Pick a concept: What role does the voice play? Sketch a quick persona.
- Prompt with intent: Describe age, accent, tone, pacing, and audio quality in a single sentence. Example: “Perfect audio quality. Elderly male, thick Scottish accent, slow and reflective, gravelly timbre.”
- Save and deploy: Choose one of the three previews—it’s instantly usable in the ElevenLabs API or web studio.
No dropdown forests. No hidden levers. Just results.
Prompting Power Tips
Below you’ll find the attributes our research team sees most often in top‑quality results:
| Attribute | Why it matters | Example keywords |
|---|---|---|
| Age | Sets vocal texture and pitch | child, teen, middle-aged, elderly |
| Accent/nationality | Grounds the character in place | thick Australian, light French, neutral American |
| Gender | Guides resonance | male, female, gender-neutral |
| Tone & emotion | Drives delivery | warm, assertive, anxious, joyful |
| Speed | Controls pacing without editing | fast, measured, languid |
| Guidance scale | Balances creativity vs. prompt fidelity | “guidance scale 10” (try 8–12 for accuracy, 3–5 for exploration) |
For a full matrix, see the Prompting guide in docs.
A word on clarity
最適なプロンプトは日常会話のように短く、具体的で、専門用語を使わないものです。この原則は私たちの執筆スタイルにも反映されています。削れる言葉は削りましょう。
Real-world applications
- Game development:今日はプレースホルダーの声でサイドクエストをプロトタイプ。明日には最終版に置き換え—再キャスティング不要。
- Audiobooks and podcasts: Give each character a distinct sound. Then refine delivery scene by scene.
- Localization: Maintain timing in translated dialogue while generating culturally accurate accents.
- Accessibility: スクリーンリーダーやIVRシステムにブランドのトーンに合った声を構築し、ユーザーの快適さを向上。
今すぐ利用可能
Voice Design v3がElevenLabsのダッシュボードで利用可能に：Voices → My Voices → Add a new voice → Voice Design
ログインし、プロンプトを入力して生成をクリック。すぐに3つの新しい声が手に入ります—そのうち1つはあなたの頭の中にしか存在しなかったものです。
スタジオでお会いしましょう。
よくある質問
いいえ。PVCは特定のライセンスされたパフォーマーを再現する際のゴールドスタンダードです。Voice Designは、その正確な声がライブラリーにない場合や新しいコンセプトを試行する際に役立ちます。
プレビューのテキストに含まれる文字数だけです—裏で3つのサンプルを生成していても。
はい。任意の属性を調整して再生成できます。新しいプレビューは同じ1クレジットルールに従います。
ElevenLabsチームによる記事をもっと見る
リソース
Eleven v3 Audio Tags: Bringing multi-character dialogue to life
Create dynamic multi-character dialogue with Eleven v3 Audio Tags. Script overlapping voices, interruptions, and emotional shifts for natural, human-like AI conversations.