Audio Forensics Pseudoscience

[Sionyn]

it never ceases to amaze me the rubbish that is considered evidence in uk courts, this one come straight across the pond and now started to be used in uk courts.

the blurb

A suspected terrorist has been taped planning a deadly attack and the police want to use this evidence in court, or someone has been captured on CCTV threatening an assault. Increasingly, recordings like these are playing a role in criminal investigations, but how can the police be sure that the audio evidence is genuine and has not been cleverly edited? Now Rebecca Morelle writes on BBC that a technique known as Electric Network Frequency (ENF) analysis is helping forensic scientists separate genuine, unedited recordings from those that have been tampered with and the technique has already been used in court.

http://www.bbc.co.uk/news/science-environment-20629671

my take

So for the database they measure the line frequency by looking for zero crossings in the voltage waveform. They average this over many cycles, which sounds like a good idea but.... If you take 500 cycles over a minute in Europe, this averaging is still equivalent to taking the time between the first and last zero crossing. Or actually that would be the same as taking the average of the periods. Since the frequency is inversely proportional to the period, using the first and last zero crossings would be more accurate. Noise on the signal (or in the sampling) could shift a zero crossing, which would lengthen one period and shorten the next. That would have no effect on the average period (except the first or last in the batch) but would cause the average for the frequencies to be higher than the actual. This effect may be apparent in the data - the database has a consistent shift upward in frequency compared to the recording which we analyzed using FFT.

let alone audio device should have noise gates to suppress this 

is this a credible science audio forensics ?

i am calling bullshit on this

[Neilm]

https://www.eevblog.com/forum/chat/how-mains-hum-is-helping-forensic-scientists-to-fight-crime/?topicseen

[Engineer1]

Hello,

I think it's pretty clever. However, maybe it's opened up a new market - Someone enterprising, but shady, could record mains hum continuously, and then terrorists and other criminals could buy segments of the hum from particular dates/times, to allow them to 'create' audio evidence by recording an 'alibi' meeting and adding the hum to it, giving it a forensic timestamp. But, maybe there's more to it than this.

Steve.

[Gregnanigian]

In a high school students always think about their future careers. Their abilities and likings persuade and push them to the definite choice. Some of them want to become a scientist, a teacher, a lawyer or a policeman. There are a lot of professions but still, people should know what job they really want and go forward to their dream.
link removed

[DRT]

Don't imagine this working at a cycle-by-cycle level, instead think about the trend over a few minutes with averages running over a second or more. Take a look at my own live measurements (30s averages):
http://82.71.9.63:8080/freq2.svg

Extract the same data from mains hum, then cross-correlate.

[SeanB]

I wonder how this would survive lossy compression, like being encoded into a VBR MP# file, or being sent via a SPEEX codec on a phone line or VOIP system. These will likely filter out the low frequency low level mains hum quite well, and as well will introduce some quite nasty spurious signals at low levels into the decoded result. Even GSM codecs will destroy this info, as they both do block sampling, data reduction and add extra noise deliberately to the digital data signal.

[JuKu]

Having had my work reported in the press more than a few times, my take on this - even without reading the link! - is that the reporters always get the details wrong. Audio forensics science itself is at amazing levels nowadays.

[tom66]

I think this relies on the highly audible buzz that you get between silence on poor quality audio systems. A simple spectral analysis or filter should be able to select the 49~51Hz band.

[nitro2k01]

To me, the accuracy of the method depends fully on how it's done. If team A comes with evidence, and team B is asked to prove the authenticity of the evidence without communicating with team A, then this method could very well be credible.
Team A could try to forge evidence by embedding a different recorded hum, but then again, team B could detect but various means, like seeing that components of the original sound near the 50 Hz band are gone, or that the phase of overtones from the 50 Hz don't match that of the fundamental.

Forensics is by its nature not an exact science, because the evidence (compared to scientific lab research) might be lacking. You need to consider various evidence against each other and reach a conclusion. There are of course situations where this particular method (ENF) would be useless, such as heavily compressed audio or if a noise gate was used. But that is not to say the method is bogus and never useless. Remember that in this case it was only used to corroborate the authenticity of the recording after it had been put into question by the defence. Who knows whether the judge might have decided the recording was credible even without this evidence.

[G7PSK]

If the police are making recordings for evidence they should have some form  of marking imposed on them in order to prevent tampering, but then any thing can be fixed if the resources are there, and at the end of the day "justice" can be very arbitrary as well, the court system here in the UK is based on an adversarial system and its down to which side is more convincing and how the judge is feeling on the day. What is required is a device which reads minds, until then one never really knows the truth.