Transcribe short speech recordings to support documentation and subsequent text processing.
Base64-encoded audio up to 5 MB and 60 seconds per request, with an optional language setting. Output: transcript and detected language. Review results before reuse.
Hourly rate: 68,077 SupraWorx Credits. Dedicated capacity is metered by AWS running time, including idle time. Shared pools meter active GPU processing time; queueing and pool idle time are platform costs. Charges are prorated and rounded up to whole credits. Each user needs a personal GPU-Hub license; compute is paid from the organization credit balance.
AWS Frankfurt integration through the SupraWorx API, Shortcuts, Workflows and AI Workbench. Model: Systran/faster-whisper-small. Review outputs before use. The pinned configuration passed real-GPU evaluation and is released. GPU allocation depends on AWS capacity. The workload measurements are documented below; availability is not guaranteed.
AWS Frankfurt, NVIDIA L4: 20 licensed English LibriSpeech test recordings; word error rate 3.23%. This test does not establish German or noisy-room speech quality. Measured p95 request latency: 461 ms; zero runtime errors in these 20 calls. Measurements cover this workload and are not a latency or availability guarantee. Released; allocation remains subject to available AWS capacity.