whisper MP4-to-srt 影片1小時以上 [mac os]

The term "Whisper CLI" typically refers to the command-line interface provided by OpenAI's open-source Whisper model for speech recognition and translation.

Installation

The official Whisper CLI is a Python package and requires ffmpeg to be installed on your system. 

1. Install ffmpeg: Use a package manager appropriate for your operating system (e.g., Homebrew on macOS, apt on Ubuntu/Debian, Chocolatey on Windows).
   bash

   # On macOS
   brew install ffmpeg
   # On Ubuntu/Debian
   sudo apt update && sudo apt install ffmpeg
   

2. Install the Whisper Python package: Use pip to install the package from the official OpenAI GitHub repository.
   bash

   pip install -U openai-whisper
   

Usage

Once installed, you can use the whisper command to transcribe audio files locally. 

• Transcribe an audio file:
  bash

  whisper audio.mp3
  

  This will automatically detect the language and transcribe the audio, saving the output as text files in the same directory.
• Specify a different model:
  bash

  whisper audio.flac --model medium
  

  Models vary in size and accuracy (tiny, base, small, medium, large).
• Translate a non-English file to English:
  bash

  whisper japanese.wav --model medium --task translate
  

  This translates the source speech into English text.
• View all options:
  bash

  whisper --help
  

   

Alternative CLIs

Several third-party projects offer enhanced CLIs, often for performance: 

• faster-whisper-cli: A Python package utilizing CTranslate2 for faster inference, especially on specific hardware.
• whisper.cpp-cli: A C/C++ port (whisper.cpp) focused on efficient, high-performance, and cross-platform on-device inference, often using CPU/GPU acceleration features like Core ML or cuBLAS. These are standalone executables and do not require Python

----------

260121

whisper apv2.mp4 --language zh --task transcribe --output_format srt

260122

whisper a.mp4 --language zh --task transcribe --output_format srt

whisper cli.mp4 --language zh --task transcribe --output_format srt

260123.   

whisper apv40.mov --language zh --task transcribe --output_format srt

260204 array AG FQDN

whisper ag.mp4 --language zh --task transcribe --output_format srt

260206

whisper m.mp4 --language zh --task transcribe --output_format srt

260211

whisper stm.mp4 --language zh --task transcribe --output_format srt

whisper cc.mp4 --language zh --task transcribe --output_format srt

260226 Array AG FQDN_CIFS

whisper cifs.mp4 --language zh --task transcribe --output_format srt

260305 ty

whisper ty.m4a --language zh --task transcribe --output_format srt

260311 max 30分鐘

whisper max.mov --language zh --task transcribe --output_format srt

輸出的 .srt 會放在 你當前的工作目錄

以上為簡體中文

簡體中文->繁體中文

brew install opencc

opencc -i a2.srt -o a2_traditional.srt -c s2t.json

opencc -i 0122.srt -o a2_traditional.srt -c s2t.json

260123

opencc -i apv40.srt -o apv40_traditional.srt -c s2t.json

260226.   CIFS Array AG

opencc -i cifs.srt -o apv40_traditional.srt -c s2t.json

如何去掉贅詞（後處理方式）

方法 1：用文字處理工具清理

1. Whisper 生成 a.srt。

2. 使用正則表達式或文字編輯器批次刪除常見贅詞。 例如在 sed 或 grep 中：

   bash

   sed -E 's/(嗯|啊|就是|那個)//g' a.srt > a_clean.srt
   

   👉 這樣會生成一個去掉贅詞的字幕檔。

方法 2：用 OpenCC + 自訂詞表

OpenCC 本來是繁簡轉換工具，但它也能套用自訂詞表。

• 你可以建立一個「贅詞刪除詞表」，讓它在轉換時自動去掉。

方法 3：用 NLP 後處理

如果要更智慧化，可以用 Python NLP 套件（spaCy、jieba）來過濾停用詞，保留主要語意。 範例：

python

import re

with open("a.srt", "r", encoding="utf-8") as f:
    text = f.read()

clean_text = re.sub(r"(嗯|啊|就是|那個)", "", text)

with open("a_clean.srt", "w", encoding="utf-8") as f:
    f.write(clean_text)

📌 彙整

• Whisper CLI → 沒有內建去贅詞功能。

• 解法 → 後處理（正則表達式 / 自訂詞表 / NLP）。

• 最快方式 → 用 sed 或 Python 腳本批次清理。