Turn a transcription into singable target-language lyrics (DeepSeek, syllable-budgeted per phrase). Pick a transcription from a previous test and a target language.