Hands-on workshops in English

Build voice apps you can explain.

Learn step by step—from the terminal to deployment. Every file is complete, every checkpoint is testable, and every architectural decision has a reason.

Choose a workshop How to follow

02 complete labs
08 executable checkpoints
EN every chapter

text → secure server audio · WebRTC · trust

Start in 5 minutes

See the outcome. Then choose how to build it.

The quick start assumes Node.js 22+, Git, an OpenAI API account with active billing, and a project API key. Voice API calls may incur charges.

01 · Recommended first

Text to Speech

Time: 2–3 h
Level: Beginner
Quick test: ~5 min

Generate, play, and download audio without exposing the API key to the browser.

Run or build

02 · Next or standalone

Realtime Voice Agent

Time: 3–4 h
Level: Intermediate
Quick test: ~5 min

Talk over WebRTC using a short-lived client secret and explicit microphone controls.

Run or build

Learning by building

Choose your voice path.

Begin with a bounded audio request and progress to a live speech-to-speech conversation.

Lab 01 · Request and response

Text to Speech

Turn validated text into expressive audio with streaming, playback, download, limits, and a secure server-side boundary.

Time: 2–3 h
Level: Beginner
Cost: Usage-based API

Open tutorial → Architecture article Finished code ↗ Starter branch ↗ First checkpoint ↗

Lab 02 · Live conversation

Realtime Voice Agent

Build a speech-to-speech agent with short-lived authorization, WebRTC, turns, mute, interruption, transcript, and safe cleanup.

Time: 3–4 h
Level: Intermediate
Cost: Usage-based API

Open tutorial → Architecture article Finished code ↗ Starter branch ↗ First checkpoint ↗

Outcome before code

See what you will build.

The Lab 01 demo is real and compressed. The Lab 02 recording is explicitly pending; no result was fabricated.

Real demonstration of the Lab 01 Text to Speech interface, from its initial screen to the audio-generation area. — **Lab 01 · Real demonstration**The interface accepts text, generates audio, and displays playback and download controls. The recording contains no credentials or personal information.

RECLab 02 · Recording pending

The space is ready, but a real Realtime conversation requires a credential, microphone, and API usage. Follow the safe plan to record without exposing tokens.

Open the recording plan →

Your pace, your starting point

Three ways to learn.

The outcome is the same. The amount of mechanical support changes with your experience and available time.

Run the finished solution

Run main, see the complete experience, and use the code for investigation.

Build from the starter branch

Recommended path: a compiling base, implementation in slices, and read-only checkpoints.

Create from zero

Create the configuration, directories, and every file manually for a long class or deeper study.

One project, two destinations

Documentation here. Voice applications on a server.

GitHub Pages hosts this static learning material. The applications use server-side routes and protected variables; deploy each lab to Vercel or another Next.js-compatible host.

View the publishing guide

Markdown + Jekyll↓GitHub Pages · workshops+Next.js host · applications