Can whisper.cpp translate to other languages besides English?

No, whisper.cpp cannot translate directly into languages other than English. The underlying OpenAI Whisper model is hardcoded and trained exclusively to handle two primary tasks: transcribing any supported language into its own native text, or translating any supported language directly into English.

If you try to use the -tr (translate) flag, it will automatically default to an English output.

To translate audio into a non-English language (for example, translating Spanish audio into Chinese subtitles), you must use a two-step workflow.

The Two-Step Translation Solution

Step 1: Transcribe the Audio to Text#

First, use whisper.cpp to output a subtitle file (.srt or .vtt) in the audio's original native language.

./main -m models/ggml-medium.bin -f spanish_audio.wav -l es -osrt

This outputs spanish_audio.wav.srt in Spanish.

Step 2: Pass the Subtitles to a Translation Model#

Because whisper.cpp cannot do X-to-Y language translation, you need to route the resulting subtitle text through a dedicated language model. You have two highly effective options to automate this:

• Option A: Local Translation (Free & Private)
  Use an open-source LLM engine like [Ollama] running a model like Llama 3 or Mistral. You can run a quick script to feed the SRT file to the local model with the prompt: "Translate this SRT file into Chinese while strictly preserving the subtitle timestamps."
• Option B: DeepL / Google Translate API
  If privacy isn't a strict constraint, you can pass the subtitle file through the [DeepL API] or a command-line tool like translate-shell. These tools are explicitly designed to translate text while keeping the .srt formatting and timestamps completely intact.