Best Model for srt

[nanni61]

What is the best Model for creating subtitles from an audio file?
Possibly in Italian

[raivisdejus]

See this https://chidiwilliams.github.io/buzz/docs/faq#4-what-model-should-i-use

As Italian is quite a large language, I would expect default Whisper models to be quite good for it. The model size will depend on tour use case. For best quality (accuracy) I would go with Large V2 or Large V3. Both are generally good, I have seen posts where people say that V3 does hallucinate a bit more, so you may get more junk in transcripts if the audio is not very clean, if you have noise in it.

[VectorPeak]

For SRTs, the “best” choice is usually the largest model your machine can run comfortably, because subtitle quality depends on both transcription accuracy and timestamp accuracy. For Italian, large-v3 is a good default if you have enough GPU/CPU time; medium is a good fallback if large-v3 is too slow.

A few practical tips for better Italian subtitles:

• Set the language explicitly to Italian instead of relying on auto-detect.
• Use transcribe, not translate, if you want Italian subtitles.
• If the audio is long, noisy, or has multiple speakers, use a larger model.
• If speed matters more than accuracy, try small or medium first and compare.

A good workflow is to run a short 2 - 3 minute sample through medium and large-v3, export both as .srt, then compare readability and timing before processing the full file.

---

如果我的回答解决了您的问题，您可以点一下 answered the question，其实我是来解答问题，顺带着收集 Galaxy Brain 徽章的哈哈 😆