Transparent Technology
Technology & Accuracy
We believe in transparency. Here's exactly what powers Zorg Subtitle, including honest benchmarks and limitations.
Our AI Stack
Speech-to-Text
OpenAI Whisper large-v3
- Trained on 680,000 hours of multilingual audio data
- State-of-the-art accuracy across 100+ languages
- Robust to accents, background noise, and technical audio
- Automatic punctuation and speaker diarization
Translation
GPT-4o / GPT-3.5-turbo
- Context-aware translation preserving meaning and tone
- Cultural nuance and idiomatic expression handling
- GPT-4o (Pro plan) for highest quality translations
- Post-processing for subtitle-optimized formatting
Accuracy Benchmarks
Honest metrics based on our internal testing. Results may vary based on audio quality.
~4-5%
Word Error Rate
Clean audio, native English speakers
~8-12%
Word Error Rate
Accented speech, non-native speakers
~15-20%
Word Error Rate
Background noise, multiple speakers
Important Note on Benchmarks
These are approximate ranges based on internal testing with varied content. Word Error Rate (WER) measures the percentage of incorrectly transcribed words. Lower is better. We always recommend human review for professional or broadcast content.
Translation Quality
| Model | Plan | Best For | Characteristics |
|---|---|---|---|
| GPT-3.5-turbo | Free & Standard | General content, social media | Fast, cost-effective, good accuracy |
| GPT-4o | Pro | Professional, nuanced content | Superior context, cultural awareness, idiomatic |
Processing Infrastructure
Cloud Processing
- Server-side transcription and translation
- No local software installation required
- Automatic scaling based on demand
- Real-time progress tracking
Security & Privacy
- Encrypted file transfer (HTTPS/TLS)
- Videos processed and not retained by AI providers
- Supabase storage with row-level security
- GDPR-conscious data handling
Ready to try it yourself?
Start with our free tier. No credit card required.