Agent assistants
Let a Discord, Telegram, browser, or support agent speak when text is too flat.
x402 voice API live on Base
Give your agent a voice.
ForgeMesh Voice turns text into speech for agents, apps, videos, books, narration, support flows, and creative tools. Pay per call. No subscription. No API key. Starts at a tenth of a cent.
$0.001base voice speech from
20voices
31languages
10persona voices
One voice layer. Many products.
Use it as the default voice for an assistant, narration for a video workflow, character lines in a game, book drafts, status updates, social clips, and audio alerts.
TikToks and shorts
Generate short narration, hooks, voiceovers, and alternate takes without a subscription seat.
Video narration
Turn scripts, explainers, walkthroughs, and product demos into clean spoken audio.
Books and stories
Draft audiobook chapters, character readings, bedtime stories, and serial narration.
Games and NPCs
Add generated lines, alerts, quest text, and system narration using persona voices.
Monitoring and ops
Turn critical changes into immediate audio alerts that a human can notice fast.
Shape how the voice delivers.
Expressions are the controls that make speech feel faster, calmer, clearer, more premium, or more urgent. Use them when the voice needs intent, not just audio.
Control pace
0.7-2.0x. Slow narration down for stories, or push agent updates a little faster. Good starting point for Assistant: 1.12x.
Control polish
steps 1-100. Use 8 for base-voice clips, 16 for polished speech, and 24 for premium narration.
Custom tier only
Assistant, Storyteller, Narrator, Announcer, Urgent, Velvet, Sage, Spark, Anchor, and Echo require Custom. They are not included in Base, Pro, Batch, or OpenAI-shaped routes.
Choose the tier
Base is 10 base voices only. Pro adds speed and quality for base voices. Custom unlocks persona voices plus speed and quality.
Recommended agent voice recipe
Start with Assistant, set speed around 1.12, and use 16 quality steps. It keeps status updates clear, warm, and quick without sounding rushed.
POST https://tts.forgemesh.io/v1/tts/custom
content-type: application/json
{
"text": "I found a new task that needs your review.",
"voice": "Assistant",
"speed": 1.12,
"steps": 16
}
Personas with purpose.
Pick the voice by job, not by a random voice id. Standard voices cover everyday speech; personas make the output feel intentional.
Storyteller
Warm, measured narration for books, podcasts, and story agents.
Velvet
Rich premium tone for luxury UX, product moments, and polished clips.
Announcer
Crisp voice for intros, news, alerts, and high-clarity updates.
Assistant
Friendly support voice for chatbots, help systems, and guided flows.
Urgent
Fast attention voice for warnings, monitoring, and incident audio.
Pricing built for agents.
Small calls stay cheap. Longer speech uses the matching bucket. Agents can probe, pay, and speak without accounts, dashboards, or monthly minimums.
Tier
1-500 chars
501-2000 chars
Best for
Base
$0.001
$0.003
10 base voices only; no speed, quality, or personas
Pro
$0.003
$0.006
10 base voices with speed control and quality steps
Custom
$0.005
$0.01
All 20 voices, including Custom-only personas like Assistant, Storyteller, Velvet, and Urgent
OpenAI-shaped
$0.001
$0.003
Drop-in request shape for the 10 base voices only
Batch
$0.002
$0.005
Up to 20 base-voice clips in one paid call; no personas
31 languages ready for speech.
Launch an English-speaking agent first, then reuse the same voice layer for global clips, product demos, stories, alerts, and support flows.
EnglishKoreanJapaneseArabicBulgarianCzechDanishGermanGreekSpanishEstonianFinnishFrenchHindiCroatianHungarianIndonesianItalianLithuanianLatvianDutchPolishPortugueseRomanianRussianSlovakSlovenianSwedishTurkishUkrainianVietnamese
The agent voice cycle.
1
Decide
Your agent chooses what needs to be said and which voice fits the moment.
2
Request
It calls the voice endpoint with text, voice, language, and expression settings.
3
Settle
The x402 challenge is paid automatically with USDC on Base.
4
Speak
The WAV returns for playback, storage, alerts, clips, or agent replies.
Setup in one URL swap.
For the 10 base voices, keep the familiar speech request shape and point it at ForgeMesh Voice. Use Custom routes when you want persona voices like Assistant, Storyteller, Velvet, or Urgent.
1
Keep the body
Send input, voice, and model fields the way your app already expects.
2
Change the host
Use tts.forgemesh.io as the speech base URL.
3
Let x402 pay
Your agent receives the payment challenge and retries with USDC on Base.
4
Play the WAV
The response is audio your app, agent, or workflow can play or store.
POST https://tts.forgemesh.io/v1/audio/speech
content-type: application/json
{
"input": "Your shipment is delayed by 18 minutes. I will keep monitoring.",
"voice": "M1",
"model": "forgemesh-voice"
}