Norwegian reference for local AI — registers, plans and honest tests.

Local speech-to-text with faster-whisper: free on CPU?

Status: worked · Date: 2026-11-02 · Cost: 0 NOK

The problem

You want to turn speech into text without sending the audio to a cloud API.

The test (tools and cost)

faster-whisper 1.2.1, model tiny, CPU int8. Short English test clip (~7 s) from a local model package. Cost: 0 NOK. No cloud API.

The result

It worked. Model load ~0.6 s, transcription ~0.5 s on the test machine. Output:

Well, I don't wish to see it anymore, observe Phoebe, turning away her eyes. It is certainly very like the old portrait.

Language detection: en (probability 1.0). This is a small model on short, clear speech. Longer recordings, dialect and overlap need a larger model (base/small) and more time — not tested in this round.

Recipe

  1. Install: pip install faster-whisper
  2. Have ffmpeg available.
  3. Run for example: 
    from faster_whisper import WhisperModel
model = WhisperModel("tiny", device="cpu", compute_type="int8")
segments, info = model.transcribe("opptak.wav", beam_size=1)
for s in segments:
    print(s.start, s.end, s.text)
  4. Start with tiny for a smoke test. Switch to small if you need better Norwegian.
  5. Keep recordings local.

Three things worth knowing

  1. tiny is fast — not the best on Norwegian dialect.
  2. The first model download takes more time than the run itself.
  3. You do not need a GPU for short clips.

Date tested: 2026-11-02

← All tests