Back to projects
Live System 2026
Fullstack Systems Engineer

Speech-to-Text Studio

A high-performance Speech-to-Text transcription and multi-language translation web application. Features client-side Web Audio API extraction (processes 1.4GB video to 16kHz WAV in <1s), an environment-aware hybrid engine (local faster-whisper CTranslate2 GPU execution locally, Groq Cloud API in production), real-time SSE line streaming, and DeepL / NLLB-200 transcript translation.

> 01. Context & Objectives

Problem Statement

Heavy video uploads choke network bandwidth and trigger storage limits, while cloud transcription services incur steep API fees. Traditional tools also lack client-side audio extraction, dynamic local/cloud engine switching, and automated multi-language translation.

> 02. Interactive Workflow Preview

Application Lifecycle Simulation

speech-to-text.simulator
AUTO PLAYING
Stage 01: Media Drop & Pre-ProcessingDrag & Drop

interview_recording_1080p.mp4 (1.4 GB)

File ingested into memory dropzone →

Heavy 1.4 GB media files are dropped into the browser for non-blocking local pre-processing.

> 03. System Architecture

Engineering Blueprint

Hybrid speech transcription architecture combining browser-side Web Audio extraction with FastAPI SSE streaming and environment-aware inference (local CTranslate2 GPU execution vs Groq Cloud API in production, plus DeepL/NLLB translation).

architecture.interactive — pipeline flow
Hover node to inspect data flow
01 // node

Web Audio Extractor

Web Audio API

02 // node

FastAPI SSE Stream API

FastAPI / SSE

03 // node

Hybrid Inference Engine

faster-whisper / Groq

04 // node

Hybrid Translation Engine

DeepL / NLLB-200

Node Inspector: Web Audio ExtractorProtocol: OfflineAudioContext (16kHz WAV)

Extracts and compresses speech audio from multi-gigabyte videos down to <45MB 16kHz WAV inside browser memory in <1s before upload.

01.Web Audio Extractor

Browser-side Web Audio API (OfflineAudioContext) downsampling multi-gigabyte media to 16kHz mono WAV in <1s before upload.

02.FastAPI SSE Stream API

FastAPI job queue manager streaming line-by-line transcription updates to React via Server-Sent Events.

03.Hybrid Inference Engine

Runs local faster-whisper (CTranslate2) on GPU in local dev, automatically switching to Groq Cloud API (Whisper Large-v3) in production.

> 04. Technology Stack

#Python#FastAPI#Whisper#React#TypeScript#Web Audio API#Groq API#DeepL API

> 05. Key Highlights

Engineering Specifications

// 01

Browser-Side Web Audio Extraction

Extracts and compresses speech audio from 1.4GB+ videos down to <45MB 16kHz WAV directly in browser memory in under 1 second.

// 02

Hybrid Engine (Local GPU / Groq Cloud)

Leverages local faster-whisper (CTranslate2) for GPU inference locally, and seamlessly switches to Groq Cloud API for production deployment.

// 03

Multi-Engine Transcript Translation

Translates completed transcriptions into multiple languages via DeepL API in production or Meta NLLB-200 locally, preserving subtitle timestamps.