The Problem
Manually transcribing audio and video recordings was slow, time-consuming, and expensive. Users also lacked an easy way to translate transcripts or review speech with accurate timestamps.
The Solution
Developed a speech-to-text platform that automatically converts uploaded audio into accurate transcripts, generates word-level timestamps, supports multilingual translation, and provides an intuitive interface for reviewing and managing transcription results.
Overview
Project Overview
VoxTranscriber Platform is a full-stack speech recognition platform that converts audio into accurate text, generates timestamps, and supports multilingual translation through a modern web interface.
System Design
System Architecture
Frontend
Backend
AI Engine
Audio Processing
Deployment
Security
Authentication Flow
User accesses the application.
User uploads an audio file.
Frontend validates the selected file.
Audio is securely sent to the FastAPI backend.
Backend processes the request and returns the transcript.
Request Lifecycle
API Request Flow
User uploads an audio file.
Frontend sends the file to the FastAPI API.
Backend validates supported audio formats.
FFmpeg prepares the audio for processing.
Vosk performs speech recognition.
Word-level timestamps are generated.
Translation is performed if requested.
Transcript is returned to the frontend.
User listens while reviewing synchronized timestamps.
Data Layer
Database Design
Uploads
- id
- filename
- file_type
- duration
- uploaded_at
Transcripts
- id
- upload_id
- language
- transcript
- created_at
WordTimestamps
- id
- transcript_id
- word
- start_time
- end_time
Translations
- id
- transcript_id
- language
- translated_text
Capabilities
Key Features
Speech Recognition
Convert spoken audio into accurate text.
Audio Upload
Upload MP3, WAV and M4A audio files.
Word Timestamps
Generate precise timestamps for every spoken word.
Multi-language Translation
Translate transcripts into multiple supported languages.
Built-in Audio Player
Listen to audio while reading synchronized transcripts.
Noise Processing
Prepare audio for improved speech recognition accuracy.
REST API
FastAPI-powered backend for transcription services.
Under The Hood
Engineering Challenges
Challenge
Speech Accuracy
Resolution
Optimized audio processing to improve recognition accuracy across different recording qualities.
Challenge
Timestamp Synchronization
Resolution
Aligned every recognized word with precise timestamps for playback synchronization.
Challenge
Large Audio Processing
Resolution
Handled lengthy audio files while maintaining responsive frontend interactions.
Challenge
Translation Workflow
Resolution
Integrated multilingual translation without interrupting the transcription pipeline.
Decision Record
Technical Decisions
Chose
Next.js
Over
React
Why
Selected for improved SEO, performance, and server-side rendering capabilities.
Chose
FastAPI
Over
ExpressJs ( because i dont know it )
Why
Selected for its high performance, asynchronous support, and automatic API documentation.
Chose
Vosk
Over
Whisper (My rejection), Because when i built this project i was using old laptop so whiper crashes my working laptop
Why
Chosen for accurate offline speech recognition with efficient processing, (Whisper excluded).
Chose
FFmpeg
Why
Used to normalize and prepare audio before transcription.
Chose
React + Vite
Over
NextJs
Why
Provided a fast and responsive frontend development experience.
Retrospective
Lessons Learned
- 01
Speech recognition quality depends heavily on audio preprocessing.
- 02
Accurate timestamps greatly improve transcript usability.
- 03
Separating transcription and translation services simplifies maintenance.
- 04
Large media uploads require efficient backend processing.
- 05
Clear API response structures make frontend integration significantly easier.
🌍 Live Project
By The Numbers
Project Statistics
Full Stack
Architecture
Vosk
Speech Engine
FastAPI
Backend
Multiple
Languages
Word-Level
Timestamping
Vercel + Render
Deployment