I suspect the translated format will be more useful to the broader readership of the journal, but you could let the editor make the final decision. I don't know the context of your figure, but it's possible if the symbols don't really have a language meaning that the original version is fine as well (i.e., if they are equivalent to labels like "A, B, C, D" rather than having some phonetic or linguistic meaning). If there is some phonetic meaning that is hard or impossible to translate, that might be even more reason to use the original, and include an explanation of the characters for English-readers.
Provide one version with your revised manuscript, and provide the editor a copy of the other version as well, explain your choice, and see what their opinion is. They might decide the translated version is sufficient (with a note that it is translated from the original), or they might want you to include the untranslated in a supplementary figure, or vice-versa.
Another standard to think about for whether to translate or not is "does this content have the same meaning when translated?" For example, if you had a questionnaire asking basic biographical information, you would expect English and Persian participants to read and participate in the same way and the translation is fine. However, if your content is in reference to phonemes or some other language construct, there might not be any direct equivalent, and information would be lost in the translation.