The Transcriber

A self-managing, GPU-accelerated transcription appliance built for high-volume audio processing at a fraction of cloud costs.

Performance

  • Throughput: 1,350 hours of audio per 24 hours (per 4090 GPU appliance). Contact us for 5090 numbers.
  • Accuracy: 86%–96% word error rate depending on configuration, comparable to leading cloud ASR services
  • Engine: Whisper-based (swappable architecture)

Cost

  • Target cost, self-hosted: ~$0.01 per hour of audio
  • Comparison: AWS Transcribe charges $0.50+ per hour
  • At 500,000 hours: ~$5,000 vs. ~$250,000 (AWS)

Architecture

  • Distributed processing pool: scales automatically as GPU hardware is added or removed
  • On-premise: runs entirely on local hardware with no internet dependency
  • Hardware: Commodity Linux PCs with NVIDIA GPUs (no proprietary hardware required)

Who This Is For

Any organization transcribing tens of thousands of hours or more per month where cloud transcription costs are a significant budget line item.

No commercial license fees. No monthly subscription costs.

On-Premise Deployment

Deploy our capture and transcription technology in your own data center. No internet dependency. No cloud costs. Email us at admin [at] radio-archive [dot] org to discuss licensing.