Web_Video2Text

Now supports videos on both Youtube and BiliBili

Input a video web page url, transcribe the video without downloading it.

• No longer using Speechmatics to transcribe audio files from online urls, but using Whisper from OpenAI as a more convinient way of transcribing. https://github.com/openai/whisper

• Add serveral easy to access command line arguments.

• For Youtube videos, yt-dlp is used to fetch video urls. More on https://github.com/yt-dlp/yt-dlp.

• For b站视频, Url parsing code is from https://www.bilibili.com/video/BV1LX4y1u7VA/?spm_id_from=333.880.my_history.page.click. and https://cloud.tencent.com/developer/article/1768680. It only supports parsing videos from BiliBili. Some amendments are made to the original code since BiliBili changed their web code.

Preparation

根据 https://github.com/openai/whisper 的指引下载 Whisper:

pip install -U openai-whisper

For Youtube:

Get yt-dlp from https://github.com/yt-dlp/yt-dlp. Guides for installation could be found on the same page, no more operation needed. This is only for fetching Youtube video urls.

For BiliBili:

在视频网页，查看网页代码F12 -> Network -> Name 并复制'user_agent'到'headers':

headers = {
    'user-agent': 'copy your user-agent here'}

以上步骤只需要进行一次。

Arguments

--model: Choose from the models provided by Whisper, tiny/base/small/medium/large. Notice this downloads any model that is not on your local device, and medium/large models are huge.

--language: Choose language to transcribe, by defualt it will be set to 'NULL', which auto-detects language. Check https://github.com/openai/whisper for supported languages.

--save: Choose whether to save the transcription as a file. It will be saved to the same directory where the code is located.

--stamp: Choose whether to include time stamps in transcribed files.

完成后直接运行并输入视频网址:

python video2text_public.py
video_url: 'your url'

The transcription will appear in terminal. 有时候抓取b站会很久都没有响应，解决方案是多跑几遍。