It may be possible to convert the speech to text first using whatever modern tools exist for it, and then feed that text as character input via PS/2. The last bit is likely gonna need a custom solution but with modern microcontrollers it is not difficult.

Unfortunately the data is stored in an SRAM with supercapacitor based supply voltage so any chance of extracting the info from old unit is probably not going to happen. It may be worthwhile to modify the new unit to replace the supercapacitor with a big lithium cell so that next time one has to worry about loss of data is in a few decades. The supercapacitor will not be able to keep the charge for very long and the data retention time after unplugging the unit is also unspecified. Everything will be lost when that capacitor (or battery) loses its charge.

Or search CDDB for the data, and then format it and transfer using a microcontroller

Seriousness:

You need a whole computer to do the speech recognition process, and then need a way to jiggle the PS/2 port's data in sync with it's clock signal to feed characters down.

This is what a pi pico, or similar, would shine at doing. It can be configured to have a bitbashed audio device on one of its gpios with a prod to the pi.conf file, and can produce the needed waveforms on its pwm capable gpios.

It might or might not be capable of being powered from the +5v line already present on the PS/2 keyboard connector, depending on the model.

If it were me, I would start googling for premade pi shields and system images for this function. It's very likely somebody has already done so.

(See for instance, pico HID. It is able to make a pico emulate a ps/2 keyboard, and do automated keyboard outputs. This is half of what you'd need, with realtime speech to text being the other. Linux speech recognition software does exist, and would be low hanging fruit.)

https://docs.pikvm.org/pico_hid/

Otherwise, you'll have to get savvy, and hack it together on your own.

Thors_Hammer wrote on 2025-12-11, 19:45:

.. a Sony 400 CD changer that died.

If it's mechanical issue, can you use the parts from second changer and fix the first one?
I assume the disc data is still in the memory of the first changer.

To to speech -> text is going to take a lot of processing power, and most likely an internet connection - most of the speech recognition stuff I've looked at use an "over the net" database.

Converting text -> PS2 is pretty easy. You can do this will a small microcontroller (At one point I made up a device to accept push button inputs from various large homemade "triggers" on the desk and enter various corresponding sequences of keys - made some games much easier!)

Thanks to all! Very crazy schedule today so just a quick reply - The "speech to text" isn't the issue as it can be done easily on PC or phone. The problem is feeding the resultant text file(s) into the PS/2 port. A microcontroller seems to be the answer, but I've limited experience with them, but am game ...! Gotta dash for the moment, check back later on any more thoughts 😀 Thx again!!