How it all works:
The Speech Synthesizer is patched in memory so that MWROBOT gets back control after the phrases have been converted to a set of parameters, giving information about pitch, three times formant frequency and level and noise frequency/level. This list is then converted to a similar list that is dumped to the Microwave. The MW list has always 61 entries and does not provide pitch information, that is the basic difference. The format of the parameter set that forms a single wave is very simple:

L1,F1,L2,F2,L3,F3,LN,FN

where

• Lx stands for the Formant #x Level in steps of 6 dB,
• Fx stands for formant frequency #x, divided by 32,
• Ln stands for Noise Level in steps of 3 dB,
• Fn stands for Noise harshness.

This list is given 61 times in the eight waves that are used by the Wavetable.

by Dr. Georg Müller from 2002:

• The MicroWave Speech
• http://12decode.de/DEADBEEF/robot_mid.zip
• Phrase List

https://youtu.be/89nv9ZOzfyU

Wavetable: User RAM I33;
Init with $DEADBEEF tag and filled with phoneme data

Wave 1: Volume = 7, Envelope Amount = 61
Wave 2: Volume = 0
Wave Envelope: Time 1 = 60, Level 1 = 127
Oscillator 1: Octave = -1

phonemelist