Skip to main content
Onur AksoyOnur Aksoy — Homepage
Production / Official Microsoft Edge Add-ons2025 — Present•Lead Architect & AI Systems Engineer

oaDub — Real-Time Multimodal AI Simultaneous Tab Dubbing Engine

An advanced Chrome/Edge Manifest V3 extension streaming 16kHz raw PCM tab audio over bidirectional WebSockets to Google Gemini Live API, synthesizing instant 24kHz voice translation with zero server intermediaries.

Latency
< 400ms Bi-directional
Audio Pipeline
16kHz PCM In → 24kHz Out
Language Support
14+ Languages
Privacy
100% Client Offline Key
#Gemini 3.5 Live#WebSocket API#Chrome Manifest V3#Offscreen Document#Web Audio API / PCM#Simultaneous Translation#Edge Add-ons

Project Overview

Language barriers on global video platforms such as YouTube, Coursera, Twitch, and web conferencing platforms hinder continuous learning and technical mastery. Traditional subtitle extensions impose continuous cognitive load, whereas legacy AI dubbing platforms require minutes of batch post-processing before rendering an alternative video track.

oaDub introduces an instant, real-time auditory translation paradigm. Built as a high-performance browser extension complying with the strict Chrome Manifest V3 lifecycle, oaDub captures live media tab audio, streams raw 16kHz Little-Endian PCM audio chunks across a low-latency bidirectional WebSocket directly to Google Gemini 3.5 Live Multimodal API, and receives streaming 24kHz PCM audio buffers. Through a client-side Web Audio API dual-channel mixer with background original volume ducking, oaDub delivers real-time simultaneous voice translation in under 400 milliseconds—all with zero intermediate servers and complete API key privacy.

System Architecture Topologyinteractive vector diagram
flowchart LR
    subgraph Browser ["Web Browser Environment (Manifest V3)"]
        subgraph Tab ["Active Video Tab (YouTube / Coursera)"]
            Video["Media Stream / Tab Audio"]
        end
        subgraph Offscreen ["Manifest V3 Offscreen Document"]
            Capture["chrome.tabCapture API"]
            AudioCtx["Web Audio API Context\n16kHz Downsampler"]
            PCM_Enc["Little-Endian PCM Encoder"]
            Capture --> AudioCtx --> PCM_Enc
        end
        subgraph ExtensionEngine ["Extension Background & Audio Mixer"]
            Mixer["Dual-Channel Audio Mixer\nOriginal Audio Ducking (-18dB)"]
            Speaker["🔊 Output Device (Speaker / Headset)"]
            Mixer --> Speaker
        end
    end
    subgraph CloudAI ["Google AI Multimodal Infrastructure"]
        GeminiLive["⚡ Gemini 3.5 Live API\nBi-directional WebSocket Endpoint"]
        SpeechSynthesis["Streaming Speech-to-Speech\n24kHz Audio Buffer Engine"]
        GeminiLive <--> SpeechSynthesis
    end
    Video --> Capture
    PCM_Enc -- "16kHz Mono PCM Frames (WebSocket)" --> GeminiLive
    SpeechSynthesis -- "24kHz Synthesized PCM Stream" --> Mixer
    style Browser fill:#08090A,stroke:#38BDF8,stroke-width:1px,color:#EDEDED
    style Offscreen fill:#0E131F,stroke:#0284C7,stroke-width:1px,color:#EDEDED
    style ExtensionEngine fill:#14171D,stroke:#38BDF8,stroke-width:1px,color:#EDEDED
    style CloudAI fill:#0B1120,stroke:#60A5FA,stroke-width:2px,color:#EDEDED

Architectural Benchmark: oaDub Live Extension vs. Traditional Dubbing Tools

Operational & Technical MetricPost-Processed Cloud DubbingStandard Subtitle ExtensionsoaDub Live Gemini Engine
Translation Latency5 – 30 Minutes (Batch Render)1.5s – 3.0s (Text STT & Translate)< 400ms (Direct Live Audio Stream)
Audio OutputPre-rendered Replacement VideoNone (Text Only / Visual Distraction)Simultaneous 24kHz Natural Voice
Server InfrastructureHeavy GPU Cloud Render FarmsTranslation API Proxy ServersZero-Server / Direct Client WebSocket
Privacy & SecurityVideo Audio Uploaded to 3rd PartyToken Logged on Remote Servers100% Client Offline Key (Zero Data Retention)
Browser IntegrationExternal Web Platform / Manual UploadStandard Content ScriptManifest V3 Offscreen Document & Tab Capture

Technical Deep-Dive

1. Manifest V3 Offscreen Document & Web Audio Tab Capture Pipeline

Modern Chrome extensions operate under Manifest V3 restrictions where background service workers lack DOM access and HTMLMediaElement lifecycle support. oaDub resolves this through an isolated Offscreen Document architecture. When enabled on any active video tab, chrome.tabCapture routes the raw media stream into a high-precision AudioContext, downsampling the multi-channel browser stream to a standardized 16,000 Hz 16-bit mono Little-Endian linear PCM stream ready for neural ingestion.

2. Low-Latency Bi-directional WebSocket & Gemini 3.5 Live Protocol

Rather than relying on disjointed Speech-to-Text (STT) followed by Text Translation (MT) and Text-to-Speech (TTS), oaDub connects directly to the Google Gemini Live Multimodal WebSocket interface. Chunked binary PCM audio frames are streamed continuously, preserving prosody and intent. Gemini Live responds with a streaming sequence of 24,000 Hz raw PCM audio chunks, achieving end-to-end voice-to-voice simultaneous translation latency below 400 milliseconds.

3. Dual-Channel Dynamic Audio Mixing & Intelligent Volume Ducking

To deliver an authentic simultaneous interpretation experience, oaDub implements an intelligent audio ducking algorithm via Web Audio API GainNodes. The original tab audio is maintained in the background at an attenuated level (-18dB) so the listener perceives natural ambient audio, while the synthesized translated voice is overlaid with pristine clarity. Dynamic circular jitter buffers smooth out variable network arrival times without introducing cumulative delay.

4. Zero-Server Privacy Architecture & Microsoft Edge Store Distribution

User privacy is guaranteed by design: oaDub features zero intermediary backend servers, zero telemetry proxies, and zero third-party databases. The user's Google AI Studio API key is encrypted and stored exclusively inside chrome.storage.local. All WebSocket handshakes occur strictly between the user's browser client and Google's official Gemini endpoint. Officially certified and published on Microsoft Edge Add-ons with automated CI/CD release pipelines.

Technical Stack

Runtime & Platform
Chrome / Microsoft Edge Extension (Manifest V3)
AI & Protocol
Google Gemini 3.5 Live API / Multimodal Bi-directional WebSocket
Audio Ingestion
Web Audio API, AudioContext, Offscreen Document, 16kHz PCM Downsampling
Output Synthesis
24kHz Low-Latency PCM Audio Stream & Dynamic Buffer Queues
Security & Privacy
chrome.storage.local Client-Only API Key Encryption
Distribution
Official Microsoft Edge Add-ons Store Certified

Business Impact & Outcomes

By eliminating intermediate proxy servers and providing direct tab audio streaming, oaDub enables barrier-free real-time global media consumption for technical professionals, students, and researchers with zero operational infrastructure cost and complete data sovereignty.

Interested in a similar system architecture?

Reach out directly to discuss technical consulting, bespoke implementation, or system audits.

Discuss Project →