Echomancer v2
Transform documents into audiobooks with stock narrators — Andrew
(default), Ava, Libby, and Ryan — plus optional voice cloning.
Live app: echomancer-v2.vercel.app
How it works
Upload or paste a book. Preview a short narrator sample. Choose a narrator.
Ready sections can play while the rest of the book generates.
Paste text if you would rather skip the file upload.
Price target: ~€4.50 for a typical take-home book. The actual quote is dynamic from length + engine (src/lib/tts/pricing.ts).
Architecture
Frontend Next.js 16 (React 19, TypeScript, Tailwind 4)
Database Turso (edge SQLite)
Storage Cloudflare R2
TTS Andrew / Ava / Libby / Ryan = Edge; user clones optional
Hosting Vercel + always-on VPS worker (Whole book)
Browser → POST /api/jobs
stream → GET /api/jobs/{id}/stream (Vercel → Edge / Fish pipe)
takehome → VM worker POST /jobs → sections → concat (podcast chain + 44.1/192; DFN opt-in) → R2 full.*
Job creation only enqueues. An always-on VPS worker (pm2)
synthesizes Whole book so a book finishes after the tab is closed.
Live Listen and Live Stream stay on Vercel. Document extract runs on the
VM worker; Cloudflare is the fallback. See WORKER.md.
Ownership
Nothing is unowned: signed-out visitors get a signed anonymous session cookie;
Google sign-in or an emailed one-time link (Resend) upgrades that cookie to a
durable user_* so a library survives devices. A job that belongs to a different session responds 404. SESSION_SECRET
is required in production; Google sign-in also needs AUTH_GOOGLE_ID and
AUTH_GOOGLE_SECRET, and email sign-in needs RESEND_API_KEY and
AUTH_EMAIL_FROM — see DEPLOYMENT.md.
Quick Start
Prerequisites
- Node.js 20+
- Turso database
- Optional: Fish API key (voice cloning, and a book already stored as Clara)
- Edge stock (Andrew / Ava / Libby / Ryan) needs no Fish or Azure key
- A book already stored as Randolph still plays and downloads. New requests use Andrew.
- Optional: R2 for production storage
Install
git clone https://github.com/joelntemuse24/Echomancer.git
cd Echomancer
npm install
Environment
TURSO_DATABASE_URL=libsql://your-db.turso.io
TURSO_AUTH_TOKEN=your-auth-token
# Signs anonymous / signed-in session cookies (required in production)
SESSION_SECRET=$(openssl rand -hex 32)
# AUTH_SECRET=... # optional; Auth.js reuses SESSION_SECRET when unset
# AUTH_GOOGLE_ID=...
# AUTH_GOOGLE_SECRET=...
# AUTH_URL=http://localhost:3000
# Fish Audio — voice cloning, and books already stored as Clara
# FISH_API_KEY=...
# Optional — leftover OpenRouter catalog ids for in-flight jobs
# OPENROUTER_API_KEY=sk-or-...
# Worker secrets
INTERNAL_JOB_SECRET=some-long-secret
CRON_SECRET=another-long-secret
# WORKER_URL=https://worker.example.com
# WORKER_SECRET=...
# Uploads (keep both in sync)
MAX_UPLOAD_MB=512
NEXT_PUBLIC_MAX_UPLOAD_MB=512
# Pricing / limits (optional)
TTS_PRICE_MARKUP=2.0
TTS_PRICE_FIXED_EUR=0.5
STREAM_MAX_AUDIO_SECONDS=3600
# Storage
R2_ACCOUNT_ID=...
R2_ACCESS_KEY_ID=...
R2_SECRET_ACCESS_KEY=...
R2_BUCKET_NAME=echomancer-audio
NEXT_PUBLIC_APP_URL=http://localhost:3000
npm run dev
The schema creates itself on first request; migrate-turso.sql does the same for
a fresh database and is safe to re-run.
Checks
npm run lint
npm run typecheck
npm run test:run
npm run build
Tests run the real route handlers against an in-memory libSQL database and a temp
storage directory, faking only the speech provider.
API sketch
| Endpoint | Purpose |
|---|
POST /api/pdf/upload | JSON presign { fileName, contentType, byteSize } — browser PUTs to R2 |
POST /api/pdf/upload/{id} | Complete after PUT; Cloudflare Worker extracts content.txt |
GET /api/pdf/upload/{id} | Poll extraction |
GET /api/tts/voices | Catalog + optional ?charCount= price estimates |
POST /api/tts/preview | One-line narrator sample |
POST /api/jobs | Create a stream or takehome job (enqueue only) |
GET /api/jobs · GET /api/jobs/{id} | The caller's library / one job |
GET /api/jobs/{id}/stream | Live listen audio |
POST /api/jobs/{id}/takehome | Spawn a full book from a stream session |
GET /api/jobs/{id}/download | Assembled audiobook |
GET /api/storage/{key} | Owner-gated blob proxy |
GET /api/cron/process-jobs | Worker drain (CRON_SECRET) |
POST /api/jobs/{id}/process | Advance one job (INTERNAL_JOB_SECRET) |
Docs
License
Private project.
