SexyVoice.ai - AI Voice Generation Platform

Generate expressive AI voices, clone your own, and talk to AI characters

🌐 Live Demo • 🗺️ Roadmap • 🚀 Quick Start • ✨ Features • 🛠️ Tech Stack • ⚙️ Deployment and release checks

🌟 About

Generate AI voices in 24+ supported languages and locales
57 public speech voices from Google Gemini (gpro, gpro31, gpro38) and xAI Grok (xai), plus 6 built-in voices for calls
Voice cloning across 28 languages
Credit-based usage system
User authentication and profile management (Google)
Architecture Overview

SexyVoice.ai brings voice generation, voice cloning, and live AI calls into one app. Generate expressive speech with Gemini and Grok voices, clone your own voice, or create a character with a personality and a voice you choose. Developers can use the external API for speech generation with the same credit balance as the dashboard. See GLOSSARY.md for domain names and definitions.

✨ Features

🎯 Core Functionality

AI Voice Generation: Create realistic voices powered by Google Gemini, xAI Grok, and additional TTS models
Voice Cloning: Clone your own voice with as little as 10 seconds of audio
Live AI Voice Calls: Talk to AI characters in real time; paid users can create characters and choose a cloned voice
Voice Selection System: Choose from 57 public speech voices: Gemini voices such as achernar, kore, puck, and zephyr on Gemini 2.5, 3.1, and 3.8, Spanish role voices for Spain and Mexico, and Grok voices such as ara, carina, eve, leo, rex, sal, and zagan
Multi-language Support: Generate speech in 24+ supported languages and locales, with broad multilingual coverage for generation, cloning, and real-time voice experiences
Audio Transcription: Transcribe audio files to text offline in 99+ languages with optional translation to English using Whisper AI

🔐 User Experience

Secure Authentication: Multiple login options with Google (more coming soon)
Credit-based System: Fair usage tracking with transparent pricing
Profile Management: Personalized dashboard and settings
Audio History: Track and manage all your generated content

🌍 Platform Features

Responsive Design: Optimized for desktop and mobile devices
International Support: Full i18n implementation powered by next-intl for global accessibility (EN/ES/DE/DA/IT/FR)
Localized Site Banners: Shared banner system for promos and announcements across landing, blog, and dashboard with independent dismiss state and one visible banner at a time
Rate Limiting: Fair usage policies to ensure platform stability
Real-time Updates: Live audio generation with progress tracking
Public Tools: Free utility tools including audio transcription and format conversion

🛠️ Tech Stack

Frontend

Next.js 16 - React framework with App Router and TypeScript
next-intl - Internationalization for Next.js App Router; messages in apps/web/messages/*.json; getMessages() for server components, useTranslations() for client components
React 19 - Server Components (RSCs), Suspense, and Server Actions
Tailwind CSS 4 - Utility-first CSS framework
shadcn/ui - Modern component library
Radix UI - Headless component primitives

Backend & Database

Supabase - Authentication and PostgreSQL database with SSR support
Drizzle ORM - Type-safe database operations (planned)
Cloudflare R2 - Scalable audio file storage with global CDN

DevOps & Monitoring

Vercel - Deployment and hosting platform
Sentry - Error tracking and performance monitoring
PostHog - Product analytics and feature flags
Axiom - Structured request logging for API routes
Stripe - Payment processing and subscription management

Development Tools

Biome - Fast linter and formatter
TypeScript - Type safety and developer experience
Contentlayer - Type-safe content management
Husky - Git hooks for automated tasks
lint-staged - Runs commands on staged files

Repository Layout

apps/web - Next.js web app deployed to Vercel.
apps/docs - Fumadocs docs app for docs.sexyvoice.ai.
scripts - operational scripts kept outside the web app as @sexyvoice/scripts.
docs - internal engineering and operational docs.

Root commands are orchestrated with Turborepo. Use package filters when you only want one app, for example pnpm --filter @sexyvoice/web dev.

🚀 Getting Started

Prerequisites

Node.js 24+
pnpm
Supabase account - https://supabase.com

Installation

Clone the repository

git clone https://github.com/gianpaj/sexyvoice.git
cd sexyvoice

Install dependencies
```
pnpm install
```
Set up environment variables
```
cp apps/web/.env.example apps/web/.env.local
```
Fill in the required environment variables as defined in apps/web/.env.example:
- Supabase
  - NEXT_PUBLIC_SUPABASE_URL
  - NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY - Safe for browser clients; database access remains controlled by RLS
  - SUPABASE_SECRET_KEY - Server-only admin key that bypasses RLS; never expose it to clients or use a NEXT_PUBLIC_ prefix
- Your Redis (Upstash)
  - KV_REST_API_URL
  - KV_REST_API_TOKEN
- Cloudflare R2 storage
  - R2_ACCESS_KEY_ID
  - R2_SECRET_ACCESS_KEY
  - R2_BUCKET_NAME
  - R2_SPEECH_API_BUCKET_NAME - Dedicated bucket for /api/v1/speech generated audio
  - R2_ENDPOINT - Your Cloudflare R2 endpoint URL (https://xxx.r2.cloudflarestorage.com)
- AI 3rd party services
  - REPLICATE_API_TOKEN - Your Replicate API token for speech generation and one-off voice cloning
  - FAL_KEY - Your fal.ai API key for reference audio enhancement
  - GOOGLE_GENERATIVE_AI_API_KEY - Your Google Generative AI API key for text-to-speech and enhance text (automatically add emotion tags)
  - XAI_API_KEY - Your xAI API key for Grok speech generation and call transcript analysis
  - XAI_API_BASE_URL - Optional xAI REST host override for the Batch API client (default https://api.x.ai)
  - CALL_ANALYSIS_REALTIME - Optional; set to true to analyse call transcripts inline in the webhook instead of via the xAI Batch API queue (emergency bypass)
- Real-time Calls (LiveKit)
  - LIVEKIT_URL
  - LIVEKIT_API_KEY
  - LIVEKIT_API_SECRET
  - LIVEKIT_AGENT_NAME - Optional; defaults to sexycall, and must match the deployed agent's registered name
- Stripe
  - STRIPE_SECRET_KEY
  - STRIPE_WEBHOOK_SECRET
  - STRIPE_TOPUP_STARTER_PRICE_ID
  - STRIPE_TOPUP_STANDARD_PRICE_ID
  - STRIPE_TOPUP_PRO_PRICE_ID
  - STRIPE_SUBSCRIPTION_STARTER_PRICE_ID
  - STRIPE_SUBSCRIPTION_STANDARD_PRICE_ID
  - STRIPE_SUBSCRIPTION_PRO_PRICE_ID
  - STRIPE_SUBSCRIPTION_FIRST_MONTH_COUPON_ID - Optional Stripe coupon applied automatically for eligible first-time subscribers
  - STRIPE_SUBSCRIPTION_FIRST_MONTH_DISCOUNT_PERCENT - Optional first-month discount percentage used to display discounted subscription pricing when the coupon is configured
- Banner and promotion configuration
  - NEXT_PUBLIC_PROMO_ENABLED - Enables promo banners and bonus-credit pricing
  - NEXT_PUBLIC_ACTIVE_PROMO_BANNER - Active promo banner id from apps/web/messages/*.json and apps/web/lib/banners/registry.ts
  - NEXT_PUBLIC_ACTIVE_ANNOUNCEMENT_BANNER - Active announcement banner id from apps/web/messages/*.json and apps/web/lib/banners/registry.ts
  - NEXT_PUBLIC_PROMO_THEME - Banner theme (pink, orange, blue)
  - NEXT_PUBLIC_PROMO_COUNTDOWN_END_DATE - Optional countdown end date for promo banners that support it
  - NEXT_PUBLIC_PROMO_ID - Promo identifier still used by Stripe metadata and credit bonus flows
  - NEXT_PUBLIC_PROMO_BONUS_STARTER
  - NEXT_PUBLIC_PROMO_BONUS_STANDARD
  - NEXT_PUBLIC_PROMO_BONUS_PRO
- Telegram cronjob
  - TELEGRAM_WEBHOOK_URL - for daily stats notifications
  - CRON_SECRET - For securing the cron API routes (/api/daily-stats, /api/call-sessions/analyze/batch) - See Managing Cron Jobs
- Axiom logging (optional)
  - AXIOM_TOKEN - Your Axiom API token for structured request logging on /api/v1/speech
- API key security
  - API_KEY_HMAC_SECRET - Secret used to HMAC-SHA256 hash API keys before storing them in the database. Generate with openssl rand -hex 32. Without this, keys fall back to plain SHA-256 (acceptable in development, never in production).
- Vercel Edge Config (optional)
  - EDGE_CONFIG - Your Vercel Edge Config connection string (automatically set when you link an Edge Config to your project)
- Additional optional variables for analytics and monitoring (Crisp, Posthog)
- E2E builds only
  - NEXT_PUBLIC_E2E_TEST_MODE - Build-time; set to true to disable the browser Sentry client so CI runs don't report into production Sentry
For environment variables, infrastructure configuration, and release verification, see Deployment and release checks.
Set up Supabase
- Create a new project at Supabase
- Run database migrations:
```
cd apps/web
supabase db push
cd ../..
```
Start the development server
```
pnpm dev
```
Open your browser The web app runs through Portless at https://sv.dev. The portless.json app name only sets the route name; PORTLESS_TLD=dev in apps/web/package.json makes the route use .dev instead of the default .localhost.

If you previously installed the Portless startup service, it may restart the default .localhost proxy on port 443 and block .dev. Remove the service before starting the app:
```
sudo portless service uninstall
sudo portless proxy stop
pnpm dev
```

Banner System

The app uses a shared banner system for both promotions and announcements:

apps/web/components/banner.tsx renders the banner UI
apps/web/lib/banners/registry.ts defines supported banners
apps/web/lib/banners/resolve-banner.ts resolves the single visible banner per placement
apps/web/components/banner.tsx writes dismissal cookies through apps/web/lib/cookies.ts

Banner copy is localized in apps/web/messages/*.json. Only one banner is shown at a time, and each banner has its own dismiss cookie.

🧪 Development

Available Scripts

Command	Description
`pnpm dev`	Start all workspace dev tasks
`pnpm --filter @sexyvoice/web dev`	Start only the web app dev server
`pnpm build`	Build workspace apps with Turbo
`pnpm test`	Run test suites
`pnpm test:db`	Run Supabase pgTAP tests locally
`pnpm test:ui`	Run Vitest UI for the web app
`pnpm lint`	Lint web and docs code with Biome
`pnpm type-check`	Run TypeScript type checking
`pnpm format`	Check web and docs formatting with Biome
`pnpm check-translations`	Validate all locale files have matching keys
`pnpm build:content`	Build web app content layer
`pnpm clean`	Check unused dependencies with Knip
`pnpm fixall`	Apply Biome format, lint, and assist fixes

Testing

Run the test suite:

pnpm test

After applying pending migrations to the local Supabase database, run the pgTAP database tests:

pnpm test:db

For the Vitest UI during development:

pnpm test:ui

Database Operations

Generate TypeScript types from Supabase Cloud Database:

pnpm run generate-supabase-types

Push schema changes to Supabase:

cd apps/web
supabase db push
cd ../..

Fetch database migrations:

cd apps/web
supabase migration fetch
cd ../..

Backup database and schema:

export SUPABASE_DB_URL=postgresql://postgres:xxx@db.yyyy.supabase.co:5432/postgres
sh ./scripts/db_backups.sh

Docs site (Fumadocs)

The docs site at docs.sexyvoice.ai is a Fumadocs (Next.js) app in apps/docs. Content lives in apps/docs/content/.

Run it locally with pnpm --filter @sexyvoice/docs dev.
Regenerate the external API reference with pnpm --filter @sexyvoice/docs generate-openapi-docs after changing the /api/v1/* request/response schemas.

Video Generation

Generate waveform videos for audio files using seewav:

pip3 install seewav
seewav your_audio.mp3 --color '0.8,0.0,0.4'

wav to mp3

# Convert WAV to MP3 with specific audio settings
# -i input.wav: Input file
# -acodec libmp3lame: Use LAME MP3 encoder
# -q:a 2: Variable bit rate quality (0=highest, 9=lowest)
# -ar 24000: Set audio sample rate to 24kHz
# -ac 1: Set audio channels to mono (1 channel)
# output.mp3: Output file
ffmpeg -i input.wav -acodec libmp3lame -q:a 2 -ar 24000 -ac 1 output.mp3

# For high quality MP3
ffmpeg -i input.wav -acodec libmp3lame -q:a 0 -ar 44100 -ac 2 output-high-quality.mp3

# For lowest quality MP3 possible
ffmpeg -i input.wav -acodec libmp3lame -q:a 9 -ar 8000 -ac 1 output-lowest.mp3

🔒 Security

SexyVoice.ai implements multiple security layers:

Authentication: Secure OAuth integration with Supabase Auth
Data Protection: Row-level security (RLS) policies in PostgreSQL
API Security: Rate limiting and request validation
File Security: Secure R2 storage with access controls
Error Handling: Comprehensive error tracking with Sentry
Environment Isolation: Separate configurations for development and production

🤝 Contributing

We welcome contributions!

Report bugs
Suggest features
Submit pull requests
Review Deployment and release checks when changing environment variables, infrastructure, or deployment configuration

Setup

Zed with Cspell extension

npm install -g cspell @cspell/dict-de-de @cspell/dict-es-es
asdf reshim nodejs
cspell link add @cspell/dict-de-de
cspell link add @cspell/dict-es-es
# restart Zed language server

📄 License

This project is licensed under the MIT License.

🔗 Links

Website: sexyvoice.ai
Roadmap: Feature requests and roadmap
Documentation: API Documentation NEW
Support: Contact support or via Chat on the Dashboard

🏗️ Project Status

SexyVoice.ai is actively developed and maintained. Check the roadmap for upcoming features and improvements.

Current Version

✅ Core voice generation functionality
✅ Voice cloning with custom audio samples
✅ User authentication and profiles
✅ Credit system and payment processing
✅ Website multi-language support (EN/ES/DE/DA/IT/FR) via next-intl
✅ Audio transcription and translation tool
✅ Real-time AI voice calls with configurable AI agents
✅ API access

Supported Voice Families and Languages

The public catalog has 63 voices: 57 text-to-speech voices and 6 voices for AI voice calls. Public voices live in the voices table, and GET /api/v1/voices returns the text-to-speech subset.

Family	Model alias	Feature	Public voices	Languages
Gemini 3.8 Flash TTS	`gpro38`	`tts`	28	Multilingual, plus `es-ES` and `es-MX`
Gemini 3.1 Flash TTS	`gpro31`	`tts`	11	Multilingual (70+ languages)
Gemini 2.5 Pro TTS	`gpro`	`tts`	11	Multilingual (24 languages)
xAI Grok	`xai`	`tts`	7	Multilingual with automatic detection
xAI Grok	`xai`	`call`	6	Multilingual

Google Gemini multilingual voices (`gpro`, `gpro31`, `gpro38`)

The same eleven Gemini voice identities exist under each Gemini model alias, so one name maps to three catalog rows. Select by voiceId, or pass the model alias together with the voice name.

Voice	Description
`achernar`	Soft
`aoede`	Breezy
`autonoe`	Bright
`callirrhoe`	Easy-going
`despina`	Smooth
`erinome`	Clear
`gacrux`	Mature
`kore`	Firm
`puck`	Upbeat
`sulafat`	Warm
`zephyr`	Bright

Gemini 2.5 Pro (gpro) voices support style prompting and the following language/locale set:

Language	BCP-47 Code	Language	BCP-47 Code
Arabic (Egyptian)	`ar-EG`	German (Germany)	`de-DE`
English (US)	`en-US`	Spanish (US)	`es-US`
French (France)	`fr-FR`	Hindi (India)	`hi-IN`
Indonesian (Indonesia)	`id-ID`	Italian (Italy)	`it-IT`
Japanese (Japan)	`ja-JP`	Korean (Korea)	`ko-KR`
Portuguese (Brazil)	`pt-BR`	Russian (Russia)	`ru-RU`
Dutch (Netherlands)	`nl-NL`	Polish (Poland)	`pl-PL`
Thai (Thailand)	`th-TH`	Turkish (Turkey)	`tr-TR`
Vietnamese (Vietnam)	`vi-VN`	Romanian (Romania)	`ro-RO`
Ukrainian (Ukraine)	`uk-UA`	Bengali (Bangladesh)	`bn-BD`
English (India)	`en-IN` & `hi-IN` bundle	Marathi (India)	`mr-IN`
Tamil (India)	`ta-IN`	Telugu (India)	`te-IN`

Gemini 3.1 Flash TTS — 70+ languages across 80+ locales (`gpro31`)

Use gpro31 only with voices returned by /api/v1/voices as model: "gpro31". Voices returned as gpro stay on Gemini 2.5 Pro.

Supports all 24 languages above, plus (non-exhaustive): Afrikaans (af-ZA), Albanian (sq-AL), Amharic (am-ET), Armenian (hy-AM), Azerbaijani (az-AZ), Basque (eu-ES), Bulgarian (bg-BG), Catalan (ca-ES), Chinese Simplified (zh-CN), Chinese Traditional (zh-TW), Croatian (hr-HR), Czech (cs-CZ), Danish (da-DK), Estonian (et-EE), Filipino/Tagalog (fil-PH), Finnish (fi-FI), Galician (gl-ES), Georgian (ka-GE), Greek (el-GR), Gujarati (gu-IN), Hebrew (he-IL), Hungarian (hu-HU), Icelandic (is-IS), Irish (ga-IE), Javanese (jv-ID), Kannada (kn-IN), Kazakh (kk-KZ), Latvian (lv-LV), Lithuanian (lt-LT), Macedonian (mk-MK), Malay (ms-MY), Maltese (mt-MT), Norwegian Bokmål (nb-NO), Persian (fa-IR), Punjabi (pa-IN), Serbian (sr-RS), Slovak (sk-SK), Slovenian (sl-SI), Swahili (sw-KE), Swedish (sv-SE), Urdu (ur-PK), Uzbek (uz-UZ), Welsh (cy-GB), Zulu (zu-ZA). Plus English, Spanish, French, Portuguese, and Arabic regional variants.

Full list: https://cloud.google.com/text-to-speech/docs/gemini-tts#language_availability

Gemini 3.1 also supports 200+ inline audio expression tags ([cheerfully], [whispering], [pause], etc.).

Gemini 3.8 Flash TTS (`gpro38`)

Use gpro38 only with voices returned by /api/v1/voices as model: "gpro38". Gemini 3.8 rows cover the eleven multilingual identities above plus 17 Spanish role voices. Each Spanish voice description records its role, accent, and pitch.

Castilian Spanish (es-ES):

Voice	Role	Type	Pitch
`es-es-advisor-8`	Authoritative Advisor	Female	Low
`es-es-assistant-3`	Digital Assistant	Male	Low
`es-es-commercial-7`	Commercial Voiceover	Female	Medium
`es-es-concierge-3`	Concierge	Female	Medium
`es-es-csagent-6`	Call Center Agent	Female	Low
`es-es-podcaster-7`	Podcaster	Male	Low
`es-es-techagent-5`	Tech Advisor	Female	Medium
`es-es-training-12`	Training Voiceover	Female	High
`es-es-tutor-12`	Tutor	Female	Medium

Mexican Spanish (es-MX, provider locale es-419):

Voice	Role	Type	Pitch
`es-419-advisor-4`	Authoritative Advisor	Female	High
`es-419-assistant-9`	Digital Assistant	Male	Medium
`es-419-commercial-1`	Commercial Voiceover	Male	Low
`es-419-concierge-5`	Concierge	Female	High
`es-419-csagent-8`	Call Center Agent	Neutral	Medium
`es-419-podcaster-1`	Podcaster	Male	Low
`es-419-training-8`	Training Voiceover	Female	Medium
`es-419-tutor-3`	Tutor	Female	Low

xAI Grok (`xai`) expressive voices

Grok text-to-speech voices:

Voice	Description
`ara`	Warm and friendly
`carina`	Breathy and enthusiastic
`eve`	Engaging and upbeat
`leo`	Authoritative and strong
`rex`	Confident and clear
`sal`	Versatile voice
`zagan`	Powerful, dramatic, and unmistakable

AI voice calls use the Grok call voices Ara, Carina, Eve, Leo, Rex, and Sal.

Custom characters can use owned saved Inworld or Mistral voices. CALL_CLONE_PROVIDER=inworld selects the provider for new call clones; set it to mistral to create Mistral profiles. Existing profiles retain their provider. Set INWORLD_API_KEY to the Basic credential exactly as Inworld shows it (already Base64; do not encode it again), with clone/write access and give the SexyCall agent access to the same Inworld workspace. Mistral profiles require MISTRAL_API_KEY in the web app and agent. An unset CALL_CLONE_PROVIDER selects Inworld as soon as this web deployment is released. Deploy the agent's inworld-tts support and configure INWORLD_API_KEY in both deployments first. To defer the switch, explicitly set CALL_CLONE_PROVIDER=mistral in Vercel before releasing the web app; set it to inworld and redeploy once the agent and credentials are ready. See saved call voices and deployment setup.

These voices support expressive inline tags like [laugh] and wrapping tags like <fast>...</fast>, plus automatic language detection and the following language/locale options:

Language / Locale	Code	Language / Locale	Code
English	`en`	Japanese	`ja`
Arabic (Egypt)	`ar-EG`	Korean	`ko`
Arabic (Saudi Arabia)	`ar-SA`	Portuguese (Brazil)	`pt-BR`
Arabic (United Arab Emirates)	`ar-AE`	Portuguese (Portugal)	`pt-PT`
Bengali	`bn`	Russian	`ru`
Chinese (Simplified)	`zh`	Spanish (Spain)	`es-ES`
French	`fr`	Spanish (Mexico)	`es-MX`
German	`de`	Turkish	`tr`
Hindi	`hi`	Vietnamese	`vi`
Indonesian	`id`	Italian	`it`

Made with ❤️ by Gianfranco

Name		Name	Last commit message	Last commit date
.agents		.agents
.claude/skills		.claude/skills
.codex		.codex
.config		.config
.github		.github
.husky		.husky
.zed		.zed
apps		apps
biome-plugins		biome-plugins
docs		docs
patches		patches
plans		plans
playwright-report		playwright-report
scripts		scripts
skills		skills
.biomeignore		.biomeignore
.ckignore		.ckignore
.envrc		.envrc
.gitignore		.gitignore
.lintstagedrc.js		.lintstagedrc.js
.markdownlint.yml		.markdownlint.yml
.npmrc		.npmrc
.prettierignore		.prettierignore
.tool-versions		.tool-versions
.worktreeinclude		.worktreeinclude
AGENTS.md		AGENTS.md
ARCHITECTURE.md		ARCHITECTURE.md
CLAUDE.md		CLAUDE.md
Changelog.md		Changelog.md
GLOSSARY.md		GLOSSARY.md
LICENSE		LICENSE
README.md		README.md
ROADMAP.md		ROADMAP.md
TODO.md		TODO.md
add-emotion-tags.prompt.yml		add-emotion-tags.prompt.yml
biome.jsonc		biome.jsonc
cspell.json		cspell.json
deno.lock		deno.lock
doctor.config.ts		doctor.config.ts
knip.jsonc		knip.jsonc
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
pnpm-workspace.yaml		pnpm-workspace.yaml
portless.json		portless.json
prompts.md		prompts.md
prompts2.md		prompts2.md
pyrightconfig.json		pyrightconfig.json
renovate.json		renovate.json
skills-lock.json		skills-lock.json
thumbs-vote.html		thumbs-vote.html
turbo.json		turbo.json

Latest commit

History

Folders and files

Repository files navigation