For circuit-switched voice, 3G/4G dongles have undocumented commands to enable an audio-only virtual serial port that sends/receives raw audio data to the host (and you control the call via good old AT commands).

For VoWiFi/VoLTE it's just SIP with custom auth (involving the SIM card, so you need to be able to talk to it - most modems provide a way to send/receive raw APDUs to the card) on a dedicated APN so you just need to implement support for this custom auth in any SIP client (Linphone?).