One of the early attempts to partially replicate human speech was the set of vowel horns made by Christian Kratzenstein in 1780. These mimicked the five basic vowel sounds without any attempt at replicating consonants.
About a decade after that, Wolfgang von Kempelen completed work on his speech engine. The final version consisted of a bellows, reed, a mock-up of the human oral and nasal cavities, a rubber mouth worked by hand, and a string-operated flap representing the tongue. This gave a fair approximation of several consonants.
Here is a video of a replica of Kempelen’s machine being operated
In 1938, Bell Labs developed a vocoder for compressing and simplifying recorded human speech. As a “proof of concept” project, they also developed a hand operated “Voder”, which could generate human speech. There were two sources of sound: a buzzing tube, for vowels, nasals and voiced consonants, and a hissing tube for voiceless consonants. These signals were passed through any of various bandpass filters.