whisper 音频文件转文字特别慢， 是否支持Java ？

[jhx1024]

两个疑问：
1.使用whisper xxx.mp3(或其他格式音频文件)特别慢。一首3分钟歌差不多要二三十分钟才能结束。

2.是否支持Java ？

[phineas-pta]

please use english

[Tex2002ans]

I translated @jhx1024 's post using DeepL:

Chinese->English:

---

whisper audio files to text is particularly slow, is Java supported? #1430

Two questions:

1. Using whisper xxx.mp3 (or other format audio file) is particularly slow. It takes about 20-30 minutes to finish a 3 minute song.

2. Does it support Java?

---

1] Whisper, on the CPU, is quite slow. It is mostly designed to go very fast using a powerful GPU.

If you want to run it faster, there is lots of discussion in this thread:

• speed up whisper? #716

where people have created other projects, like:

• faster-whisper

which run with less RAM and can achieve similar results in a fraction of the time.

If you can only run CPU, you may also be limited to some of the smaller models:

• tiny
• base
• small
• medium

Trying to run large-v2 will be MUCH, MUCH, slower.

2] No. Not that I'm aware of. Whisper normally runs in:

• Python

and there are a few other projects that take it and wrap it in C++ or Java (for Android/iOS ports). See some of the discussion here:

• https://github.com/openai/whisper/discussions?discussions_q=is%3Aopen+android
• https://github.com/openai/whisper/discussions?discussions_q=is%3Aopen+ios

[jhx1024]

Thank you for answering my question.