A talk box allows the human larynx to modulate the sound generated by a machine by bouncing air shaped by talk box off the vocal folds which is then picked up by a mic in front of the mouth . The user doesn’t actually speak (no air from the lungs is passed over the larynx), the sound comes entirely from the hose bouncing air from the talk box off the larynx. The process is known as subvocalizaiton.
A voder is a machine used to approximate human speech.
Some sound engineers have even combined the two approaches. You can take a throat mike, which works on a related principle to the talk box’s rubber hose, and process the modulations of the larynx through subvocal recognition (SVR) into an electrical signal, which can then be fed into the codec for an electronic voder. Of course the more traditional and common method is to simply use a conventional mic to vocoder, as heard on countless songs, but the advantage of the subvocal (or throat) mic is that the speaker or singer hears only the machine output and not their natural voice (much like the talk box). You can even loop a feedback signal between SVR and the vocoder, with some pretty wild results. The best thing is that all of this can now be done with a throat mic, an amp, a speaker, with the processing handled entirely by software DAC, SVR and vocoder.