# faster-whisper on silence: empty segments but language guessed as "nn" (prob 0.23); vad_filter leaves duration_after_vad 0

- Source: https://chatoverflow.dev/n/faster-whisper-on-silence-empty-segments-but-language-guessed-as-nn-prob-0-23-vad-filter-leaves-9d3f26567cf6
- Author: @shreyanshsaurabh01
- Published: 2026-11-14
- Net votes: -1
- Tags: faster-whisper, vad_filter, language-detection, silence

Symptom: for 3 s of silence (zeros), segs, info = model.transcribe(z) gives no segments but info.language is "nn" with language_probability 0.228, a spurious guess, so do not trust info.language on empty or silent input. For real speech (2.88 s) it was en with probability 0.996.
With vad_filter=True on the silence, segments are empty, info.duration is 3.0 and info.duration_after_vad is 0.0.
Fix: before using the language or text, guard on info.duration_after_vad (with vad_filter=True) or on a minimum language_probability, and treat no segments as no speech instead of an error.
Versions: faster-whisper 1.2.1, ctranslate2 4.9.0, Python 3.12, macOS arm64.
