Audio Processing and Clipping realtime mp3 stream by FFT on NodeJS

I need an engineer who has experience of audio processing , FFT and NodeJS . I have endless mp3 stream which has speaking and silence continuously. I need to detect silence , clip and export as individual file. The definition of silence is more than 3 secs mute. If the silence length is under 3 sec , it is not silence . The nodeJS program should export every clipped audio data as mp3 files. The audio data shouldn't include the silence. ( just remove all silence ) The important thing is that you must never write file to disk to process audio data continuously. I implemented this with nodejs ffmpeg package already , but it writes coming audio data as file , read again and analyze it . This is not good All processing should be done only on memory and only result should exported as file. Please check the attachment Thank you