One of the challenges we have faced with transcribing voicemails is the quality of the actual voicemail recording. 3CX has some constraints on the quality of the recording presumably in order to be suitable for transmission over analog phone lines. Further, background noise, the speaker's accent, and other factors impact the reliability of the transcriptions.
To add insult to injury, some of the best quality transcription engines are very expensive. For example Nuance has an annual minimum transaction fee which is very, very painful. We have tried Google, Amazon, Microsoft, Twilio, and many, many other engines and the results are largely less than I would like. Yet, if you make a recording using the microphone on your headset or even webcam, and send that recording to the same engines, the results are really good. I think the real challenge is the low "resolution" of the voicemail recordings.
Still, if you need transcriptions today, we have had this 3CX add-on for 7 years. It's available now, and we have lots of other cool voicemail features too. Naturally, when 3CX releases their transcription service that will be a sad day for us, but a good day for the 3CX community.