Would it be possible to add more information to the dictation cards like in flashcards, where the user picks what to show? Sometimes, it’s quite hard to understand which exact word is asked for. Either due to unclear pronunciation or because of homophones (which very often happens, for instance, in Japanese). Also in Japanese, I keep forgetting whether I’ve added a word with kanji or kana, especially if the two are used more or less interchangeably. For now, I try to stick to kanji, but it’s not always convenient.
If the idea of dictation is to stick to audio input alone, one could just add a button that triggers reading of the context, which already be very helpful.