The Spektrum could speak. It required machine code but it was more of a rasp than speach. I persevered with machine code and managed to get the Spektrum to record and analyse audio. So I recorded two seconds of audio, sought out the quiet period then the noise and tried to match it against pre recorded words. It was like shouting at a deaf uncle, sometimes it responded, most often incorrectly!
I was only analysing the amplitude, not frequency, and getting the lowest variance from half a dozen pre recorded templates. I remember my Dad being mightily unimpressed at its performance! Bit like asking my Audi to do something now!