Speech-to-text

Whisper small requirements, license & speed

By OpenAI · released 2022-09-26 · 242M parameters

Commercial useYes
Min memory~2 GB VRAM
Runs on CPUYes
StreamingNo

Key facts

License
apache-2.0 source
Commercial use
Yes source
Apache-2.0 on HF card; MIT on GitHub
Parameters
242M source
safetensors parameter count
Minimum memory
~2 GB VRAM source
Runs on CPU
Yes source
faster-whisper README benchmarks the small model on CPU (Intel Core i7-12700K, 8 threads)
Streaming
No source
30-second window model; streaming only via wrappers
Voice cloning
n/a (speech-to-text)
Languages
en, zh, de, es, ru, ko, fr, ja, pt, tr, pl, ca, nl, ar, sv, it, id, hi, fi, vi, he, uk, el, ms, cs, ro, da, hu, ta, no, th, ur, hr, bg, lt, la, mi, ml, cy, sk, te, fa, lv, bn, sr, az, sl, kn, et, mk, br, eu, is, hy, ne, mn, bs, kk, sq, sw, gl, mr, pa, si, km, sn, yo, so, af, oc, ka, be, tg, sd, gu, am, yi, lo, uz, fo, ht, ps, tk, nn, mt, sa, lb, my, bo, tl, mg, as, tt, haw, ln, ha, ba, jw, su source
Released
2022-09-26 source
Hugging Face repo creation date (proxy for release date)
Gated on HF
No

Measured speed & memory

HardwareRuntime / settingsRTFFirst-segment latencyPeak RAMWER*Date
Cloud x86 VM, 8 vCPU Intel Xeon, 16 GB RAM, no GPUfaster-whisper small (int8, beam 5)0.1008 (10× real time) 1.017 s989.8 MB 0.0%2026-10-21
Cloud x86 VM, 8 vCPU Intel Xeon, 16 GB RAM, no GPUwhisper.cpp small (ggml fp16, beam 5, 8 threads, pywhispercpp)0.2958 (3× real time) 2.984 s800.9 MB 0.0%2026-10-21

Median of 5 runs after one warm-up, ~10 s English passage. Hardware is labelled exactly: a cloud x86 VM is not an N100 or a Pi. Full method on the CPU only page. Peak RAM is whole-process RSS. *WER on clean synthetic speech, a sanity check, not a real-world accuracy figure. Methodology

Accuracy

We don't run accuracy evals. For word error rate, check the Hugging Face Open ASR Leaderboard.

Comparisons: faster-whisper vs whisper.cpp

Similar STT models

Spotted an error? Report it. Every fact links to the model card or repo it came from.

New open speech models: will they run on your hardware?

One email a week: new open TTS/STT models, license gotchas, and what we measured on CPUs, Macs, Pis and consumer GPUs.