sebastjan/reels-app
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix: handle Whisper transcribe failure for instrumental-only audio (fallback to empty transcript)

Browse Source
This commit is contained in:
Sebastjan Artič 2026-04-29 06:33:52 +00:00
parent 33a138af9e
commit e072eec362
1 changed files with 21 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
scripts/analyze.py
View File

@ -47,19 +47,31 @@ def extract_audio(video_path):
def transcribe_full(audio_path, lang=None, model_size="small"): def transcribe_full(audio_path, lang=None, model_size="small"):
"""Whisper transcript celega avdia. lang=None → auto-detect.""" """Whisper transcript celega avdia. lang=None → auto-detect.
Vrne empty transcript če Whisper ne najde govora (popolnoma instrumental)."""
from faster_whisper import WhisperModel from faster_whisper import WhisperModel
print(f"🧠 Whisper {model_size}, lang={lang or 'auto'}", file=sys.stderr) print(f"🧠 Whisper {model_size}, lang={lang or 'auto'}", file=sys.stderr)
m = WhisperModel(model_size, device="cpu", compute_type="int8") m = WhisperModel(model_size, device="cpu", compute_type="int8")
segs, info = m.transcribe( try:
audio_path, segs, info = m.transcribe(
language=lang, audio_path,
word_timestamps=True, language=lang,
vad_filter=True, word_timestamps=True,
) vad_filter=True,
detected_lang = info.language )
detected_prob = float(info.language_probability) detected_lang = info.language
detected_prob = float(info.language_probability)
except (ValueError, RuntimeError) as e:
# Whisper failure (např. pri popolnoma instrumentalnih datotekah z VAD)
print(f" ⚠️ Whisper transcribe failed: {e}", file=sys.stderr)
return {
"language": "unknown",
"language_probability": 0.0,
"segments": [],
}
print(f" Detekcija: {detected_lang} (p={detected_prob:.2f})", file=sys.stderr) print(f" Detekcija: {detected_lang} (p={detected_prob:.2f})", file=sys.stderr)
segments = [] segments = []