# How to distinguish between recorded and a live vocal signals?

Is it possible to detect if a sample of human voice by smartphone/computer(44Khz) is sourced from another electrical device or originated from a human being

I want to build a model in matlab such that, samples a vocal signal and decides whether it is a real human being talks a live or recorded, is it even possible?

• What are you trying to do? Voice-activated door lock or something? If a person is talking "live" through Skype, should that be disqualified? What you're really testing is whether a person is actually talking in front of the mic or if their voice is being played through a speaker? – endolith Oct 17 '13 at 19:29
• @endolith Yes I try to test whether a person is actually talking in front of the mic or if their voice is being played through a speaker. – 0x90 Oct 17 '13 at 20:10
• I would focus on the properties of the speaker, then, not the microphone. Portable/phone speakers are generally limited in bandwidth and all speakers have at least a little distortion. Maybe you could detect intermodulation distortion of the voice or something? – endolith Oct 17 '13 at 21:44

On a smartphone, flat out no. The microphones are so poor that you're lucky to get any approximation of the actual audio.

On high-quality hardware, perhaps. Barely. If you're lucky. Check for MP3 compression artifacts. If the source is not compressed with a lossy codec, you're out of luck.

• I can assume the recorded is compressed, you suggest trying to detect some artifacts? – 0x90 Oct 14 '13 at 3:54
• Stationary background noises will continue to be present, by definition, but speech will mask them. Perceptual compression algorithms like MP3 will discard those frequency components. So if you suddenly see the background noise disappear, you can suspect compression and thus a recording. – MSalters Oct 14 '13 at 8:52
• I disagree with the statement about smart phone microphone quality. These days most better phones use PCB mounted MEMS microphones from companies like Knowles, ST, Analog Devices, etc. which are really quite good if properly mounted. It perfectly is possible to do decent recordings with these mics, if proper software is used. – Hilmar Oct 14 '13 at 17:03
• @Hilmar: What do you think their frequency response is above 5 kHz, for instance? – MSalters Oct 14 '13 at 17:10
• A good MEMS is flat to about 10kHz (within one dB or so) and then rises because of the mechanical resonance. The rise is very consistent so it can easily be EQ'd out, if that's a problem. – Hilmar Oct 15 '13 at 1:43