范式转变
从内容生产到交互体验
第一波生成式 AI 改变了内容生产。文本、图像、视频和语音都可以按需生成，一个 prompt 就能把想法变成内容。但这些内容大多仍是一次性生成、随后被消费的成品。下一波浪潮带来的是交互性：AI 角色、游戏、陪伴应用、互动故事和语音 agent 创造的体验实时展开，并随每个用户、每个上下文、每个瞬间而变化。
内容开始持续回应正在体验它的人。
这一范式转变让文本转语音的要求不再止于自然度。自然的语音仍是高质量声音体验的基础，而交互体验带来了三个新的要求：
多样性
让每个角色拥有独一无二的声音。
可控性
随意图与上下文的变化调整每一次演绎。
低延迟
保持交互本来的自然节奏。
这些能力结合在一起，让声音可以为角色而设计、为当下而执导、并实时呈现。
全新模型
Breeze TTS 2：为交互体验而生的全新语音模型。
Breeze TTS 2 把自然语言音色设计、面向任意音色的指令遵循，以及低延迟流式生成融合进同一个模型：设计任何音色，掌控每一次演绎，并实时呈现。
推进这些能力，也需要更好的衡量方式。TTS 评测长期以自然度为中心，音色设计、指令遵循和延迟一直缺乏系统化评估。Breeze TTS 2 将这些能力汇聚在同一个面向交互式声音体验的模型中，让创作者和开发者能更精细地控制音色如何被设计、演绎和实时交付。为了推动这些领域持续进步，我们同步开源了音色设计、指令遵循与延迟三个基准测试。接下来，让我们逐一看看定义 Breeze TTS 2 的四项能力：音色设计、指令遵循、低延迟与多语言语音。
能力 01
音色设计：为每个角色找到合适的声音
从一位主角到成千上万的 NPC，Breeze TTS 2 让每个角色都拥有自己的声音。
Breeze TTS 2
MiMo-V2.5-TTS
Inworld Voice Design
Fun-AudioGen-VD
Eleven v3
Qwen3-TTS-VD
VoxCPM2Breeze TTS 2 在 TTS 音色设计基准测试中排名第一：角色契合度得分最高（78.02），音色多样性最高（708），文本通过率达 98.5%。其角色契合度领先第二名 5.24 分，可生成的独特音色数量比最接近的竞品多 39%。
Animation & Comics
PROMPT
Masculine, mature adult. Massive heavy-set alien. Deep, gravelly, resonant bass-baritone. Gruff, authoritative, commanding, drill-sergeant cadence. Fast, high-intensity. Rough, weathered, immense power, stern mentorship.
SCRIPT
Chest up, eyes forward. Space does not care if you're tired, scared, or homesick; it only rewards discipline. Stay on my count, recruit, and I'll drag a champion out of you yet.
Film & TV
PROMPT
Young adult female, mid-20s, American accent. Professional broadcast journalist turning frantic. Clear, articulate but strained by panic. High tension, rapid pace, breathless. Polished to raw desperation.
SCRIPT
This is Elena Vance, reporting live from the east block where the crowd just broke the barricade—if anyone can hear me, we need medical teams now, please, they're getting closer.
Games
PROMPT
Female, young adult, ethereal trickster. High-pitched, breathy, mischievous. Fast, rhythmic delivery with giggles. Mocking sing-song riddles. Playful, taunting, otherworldly.
SCRIPT
Oh, you heard the whisper too? Clever little heartbeat. Follow the violet shimmer if you dare, but mind your shadow, love. It tends to trade secrets when the veil grows thin.
Literature & Stage
PROMPT
Male, mid-20s to early 30s. Smooth velvety tenor-range speaking voice with a light rasp. Soulful and deeply emotive, shifting from weary struggle to optimistic conviction. Impeccable diction, gentle Southern lilt, deliberate pace, and broad expressive pitch arcs.
SCRIPT
I've walked through too many long nights to mistake silence for peace, so when I speak of tomorrow, believe me, brother, I mean a dawn we can finally stand in together.
Myth & Folklore
PROMPT
Mature nomadic storyteller. Weathered, resonant speaking voice with rhythmic pacing and expressive pitch contours. Wide dynamic range from hushed whispers to booming proclamations; soulful.
SCRIPT
Come closer to the fire, child; the wind is telling old truths tonight. Hear how it bends the grass, then remember this: every road you take is also taking measure of you.
Research & News
PROMPT
Mature male, 50s-60s, Japanese-accented English. Dignified, measured, stoic tone. Clear texture with slight age rasp. Deliberate pacing, significant pauses. Conveys wisdom, authority, and visionary caution.
SCRIPT
We do not rebuild a nation with noise or haste. We place each stone with care, and when doubt rises, we answer it with discipline, patience, and an unshaken sense of duty.
Web & Periodical
PROMPT
Ancient male, massive scale. Deep, resonant bass, gravelly, subterranean. Slow, deliberate, aristocratic. Terrifying power, sibilant, predatory intelligence.
SCRIPT
Careful where you place your little hands, treasure-seeker; every coin in this vault knows its master, and I am not inclined to forgive even the smallest theft.
能力 02
指令遵循：为每个瞬间呈现合适的演绎
用自然语言指令和内联标记塑造情绪、意图、语速与表达方式，同时保持每个角色的辨识度。
Breeze TTS 2
MiMo-v2.5-TTS
StepAudio 2.5 TTS
Qwen-Audio 3.0-TTS-Plus
Inworld TTS-2
VoxCPM2Breeze TTS 2 取得最高的指令遵循得分（4.25），领先第二名 13%。结合 0.67 的 SIM 得分，这一结果表明它能在情绪、意图、语速与表达方式上遵循自然语言指令，同时让所选音色在不同演绎中保持辨识度。
Communicative Intent
PROMPT
Speak with an alluring, hushed intensity, using a hypnotic and persuasive rhythm to draw the listener in and convince them to acquire a rare item.
SCRIPT
You feel the energy radiating from it, don't you? This isn't just a piece of quartz. It's a conduit. For a small offering, you can take it home and finally clear that dark cloud hanging over your future. Trust me, you need this.
REFERENCE
OUTPUT
Composition
PROMPT
Speaking while heavily out of breath and gasping for air, trying to reassure the listener that everything is fine.
SCRIPT
[gasps] I'm okay... [pants] just give me a second to catch my breath. [gasps] I promise, I'm completely fine... just ran a little too fast.
REFERENCE
OUTPUT
Emotion
PROMPT
The speaker sounds acerbic and bitter, delivering the lines with heavy, passive-aggressive sarcasm.
SCRIPT
Enjoy your vacation while I stay here and clean up the mess you left behind for me. Have a wonderful time.
REFERENCE
OUTPUT
Role
PROMPT
Speak with a gritty, swaggering, and theatrical tone, projecting an aggressive and boastful attitude of a seafaring marauder.
SCRIPT
Drop the anchor and bring out the gold! If any of you scallywags think about crossing me, you'll be swimming with the sharks before sunset. This ocean belongs to me!
REFERENCE
OUTPUT
Variation
PROMPT
Start with a confident, reassuring tone. At the word 'Wait', suddenly shift to terrified, breathless alarm as if reacting to a frightening noise.
SCRIPT
There's absolutely nothing to worry about, I've done this a hundred times before. Wait... did you hear that sound? Something's definitely not right here.
REFERENCE
OUTPUT
能力 03
低延迟：以对话的速度生成语音
Breeze TTS 2 将自然度、音色设计与指令遵循融入一个为实时交互打造的低延迟模型。
Breeze TTS 2 在 TTS Latency Benchmark 上排名第一，在 p50 与 p95 两个分位上都拥有最快的首音频时间。稳定的低延迟让实时交互始终灵敏，保持对话本来的自然节奏。
实时流式生成 Breeze TTS 2
Realtime API 在多个对话轮次间保持同一条 WebSocket 会话：文本一边追加，模型一边生成，PCM 音频随生成实时流回。
import { BreezeBlueClient } from "@breeze.blue/sdk";
const client = new BreezeBlueClient();
// One WebSocket session, many conversation turns.
const connection = await client.textToSpeech.realtime.connect("voc_...", {
modelId: "breeze-tts-2",
});
const consumer = (async () => {
for await (const message of connection) {
if (message.type === "audio") {
play(message.audio); // raw PCM, streamed while the model generates
} else if (message.type === "turn.done") {
return;
}
}
})();
connection.startTurn("turn_1");
connection.appendText("Hello from Breeze TTS 2.");
connection.flush();
connection.endTurn();
await consumer;
connection.close();能力 04
多语言语音
以 50 种语言生成自然、富有表现力的语音，并通过口音控制适配不同地区与受众。
选择语言
Every journey finds its meaning when someone dares to take the first step.
原文: Every journey finds its meaning when someone dares to take the first step.
