Automatic Speech Recognition

Speech Recognition that runs on CPU.

8.9 concurrent streams per vCPU at RTF under 1.05. French at 4.2% WER. 40+ languages with European depth. No GPU required — ever.

Production-grade8.9 streams/vCPURTF <1.054.2% WER FrenchCPU-only
Capabilities

Accuracy and efficiency, measured and published.

8.9 streams per vCPU

CPU-only inference at real-time factor under 1.05. Roughly 3.5× the CPU density of Speechmatics, with no GPU requirement for real-time work.

French at 4.2% WER

Top-tier accuracy on French, including regional accents, spontaneous speech, and phone-quality audio. Benchmarked monthly and published publicly.

Sub-200ms first-audio latency

Streaming ASR surfaces words as they are recognized, not at the end of an utterance. P95 first-audio under 200ms on production traffic.

40+ languages, European depth

German, Spanish, Italian, Portuguese, Dutch, Polish, Czech, Hungarian, and growing. Each language is production-grade and benchmarked publicly.

Integration

One endpoint, three deployment modes.

curl https://api.voxist.com/v1/transcribe \
  -H "Authorization: Bearer $VOXIST_API_KEY" \
  -F audio=@call.wav \
  -F language=fr \
  -F diarize=true

The /v1/transcribe endpoint supports streaming and batch mode, with diarization, word-level timestamps, and confidence scores. Available as European SaaS, private cloud, or on-premise.