openai-whisper-cli

name	openai-whisper-cli
description	Command-line usage for OpenAI Whisper, focusing on audio transcription, translation, and specifying models or languages. You MUST load this skill when interacting with the whisper command.
license	MIT

openai-whisper-cli

WHEN TO USE

• Transcribing audio files into English or other languages using whisper.
• Translating non-English speech into English using whisper --task translate.
• Selecting specific Whisper models (e.g., turbo, medium, large) for speed vs accuracy tradeoffs.
• Specifying the source language for transcription using --language.

WHEN NOT TO USE

• Using OpenAI Whisper from within Python scripts (this skill focuses purely on CLI usage).
• API integrations with OpenAI's hosted Whisper endpoints (this is for local execution).

Core Process

1. Model Selection: Determine the right model. The default is turbo, which is fast and good for English transcription but not for translation. For translating non-English to English, use medium or large.
2. Language Specification: If the audio is not in English, it is often helpful to specify the language with --language.
3. Task Definition: Use --task translate if you need the output in English from a non-English audio file.
4. Execution: Run the whisper command with the appropriate flags and input audio file.

Command-Line Examples

• Basic Transcription (Default turbo model, good for English):

  whisper audio.flac audio.mp3 audio.wav --model turbo
  

• Transcribe non-English speech (e.g., Japanese):

  whisper japanese.wav --language Japanese
  

• Translate non-English speech into English:

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

  (Note: The turbo model will return the original language even if --task translate is specified. Use medium or large for best translation results).

Core Principles

• Model Capabilities: The turbo model is optimized for transcription speed but degrades for translation. Always switch to medium or large when the goal is translation to English.
• Multilingual vs English-only: .en models (like base.en, small.en) tend to perform better for English-only tasks. Without .en (like base, small), the model is multilingual.

Common Pitfalls

• Using turbo for Translation: Pitfall: Using whisper --model turbo --task translate. Prevention: The turbo model is not trained for translation tasks. Switch to --model medium or --model large.
• Missing Dependencies: Pitfall: whisper fails to run due to missing ffmpeg or Rust. Prevention: Ensure ffmpeg is installed via system package manager (e.g., apt install ffmpeg, brew install ffmpeg).

References

• OpenAI Whisper Repository
• OpenAI Whisper Discussions USE FOR: searching issues or answers to questions.
• OpenAI Whisper Pull Requests USE FOR: searching issues or features.