Audio forensic problem - detection of time stretching - analog to digital to analog

[Tony1]

Hi, I have an audio forensic problem I am trying to resolve.

I have some analog cassette tapes that are fakes that I suspect were digitised, time stretched and then remade to tape. Their length is fixed due to the recording equipment used, so if time stretching could be proven it would identify those responsible in the chain of custody.

If the likely technique could be identified it would greatly assist me in approaching someone with the particular skill.

 

[staticV3]

I think without some reference sound inside the recording to compare against, the stretching would be difficult to prove.

Does the recording contain mains hum? Even very low levels should work. If so, what's the frequency?

Does it match the mains frequency of the location where it was recorded?

 

[Tony1]

Hi, thank you for responding. Yes there is a 50Hz hum present which is to be expected in the UK.

 

[staticV3]

All I can think of then is to search the recording for other reference sounds, things like ringtones or chimes that could be traced back and compared against the original (duration and pitch), or voices from known persons (pitch).

Another possibility is the sound of certain events (closing doors etc).
If you can identify and timestamp two of them, then you can compare the physical time difference to the recorded time difference.

Taking a look at the spectrum of the noise floor of the tampered tape and an original tape at the same location recorded at a similar time could be worth a shot as well.

 

[Redacted]

 

[Tony1]

I have done all of the normal things and am now looking for something more electronic, which is why I came on this forum. Measuring quantization artifacts has been suggested but exceeds my skill level.

 

[MaxwellsEq]

Interesting problem. Your suspicion is that speed modification happened in the digital domain.

Presumably you are confident that it wasn't just simply "slowed down" liked an old-fashioned speed control since that would result in lowered frequencies (you would detect it because pure 50Hz mains signatures would become e.g. pure 48Hz signatures).

if something else has been done, you may be able to look at the 50Hz waveforms - are there impurities or glitches. Digital time delay might be done with sample repetition and or/interpolation.

The digital to analogue conversation might show signs of this, but equally, since it's conversion is back to cassette, the latter's poor high frequency performance, noise and speed stability issues may well mask and smooth anything subtle being done in the digital domain.

 

[kemmler3D]

Time stretching that I'm familiar with will either use a spectral (FFT-based), windowed, or combination algorithm. They're getting pretty good these days but if you look carefully using a spectrogram, a time stretched piece of audio will tend to show partials that fade in and out more frequently than they would naturally, evidence of repeated content, or things like frequency quantisation of inharmonic content.

Time stretching is pretty good these days and if it's hard to hear it's probably hard to see in the analyzer.

If it's not time stretched per se and just slowed down or sped up in the same way tape is, then the digitisation won't show any telltale signs unless the change was extreme, in which case you might see missing noise spectrum at the top or bottom of the frequency range.

I should point out I'm NOT an audio forensics person, just a guy who has spent time playing around with audio, so take my thoughts for what they are, idle commentary from someone on the internet. If this is an important matter, you should seek a professional.

 

[Waveform Fidelity]

Maybe you could look at the wow and flutter?

 

[somebodyelse]

Does the recording contain mains hum? Even very low levels should work. If so, what's the frequency?

Does it match the mains frequency of the location where it was recorded?

That reminds me of something I read about a while back - people looking at keeping a record of the variations in mains frequency over time in different locations, and using that to match mains noise in a recording to try to identify time and location. I don't know whether it proved viable or not, and analog speed variations might complicate it somewhat.

 

[Mnyb]

Mains frequency variations are very small but they do vary over a 24hr period in almost every country there is some authority responsible for the mains frequency to keep it correct and another practice was ( I don’t know if it’s done anymore ) to keep the average very close to 50Hz over 24 hr period, you compensated for the dips with some deliberate peak , there where time keeping devices that used the mains as a reference way back .

The variations are very small don’t know if a tape can preserve them ?

 

[Guermantes]

Electrical Network Frequency (ENF) analysis is certainly something that has been done in the UK: https://en.wikipedia.org/wiki/Electrical_network_frequency_analysis

Ostensibly how old are the recordings in question? Are they claimed to be decades old because of the format? That may limit the availability of mains frequency records.

An old technique for cassette tape authentication is magnetic development which involves depositing fine iron particles on the tape surface to reveal record and erase head signatures with Bitter patterns. Duplication or anomalies in these signatures can point to copying or other tampering. If you have access to the alleged recorder you can see if the erase-record head gap marks on the tape match that of the recorder itself. This process can be somewhat damaging to the tape, however. I've seen that someone has looked into replaying tapes at high speed and doing spectral analysis of the low-frequency elements:

https://www.researchgate.net/publication/324228089_Forensic_audio_authentication_analysis_technique_of_first_or_higher_generation_copies_of_analog_magnetic_audio_tapes

A pity there isn't a ticking clock in the background! Failing anything obvious like clocks or music, you might be able to measure air conditioning hum, phones ringing, etc., to see if the cadences are not correct.

 

[Ruediger]

I have some analog cassette tapes that are fakes that I suspect were digitised, time stretched and then remade to tape. Their length is fixed due to the recording equipment used, so if time stretching could be proven it would identify those responsible in the chain of custody.

Could you record a sample, like a whole song, get the equivalent in original and cut/put those in an editor with timecode?

 

[somebodyelse]

Mains frequency variations are very small but they do vary over a 24hr period in almost every country there is some authority responsible for the mains frequency to keep it correct and another practice was ( I don’t know if it’s done anymore ) to keep the average very close to 50Hz over 24 hr period, you compensated for the dips with some deliberate peak , there where time keeping devices that used the mains as a reference way back .

The variations are very small don’t know if a tape can preserve them ?

They're bigger and sometimes faster than I expected - I went looking when someone said it could be audible as wow in turntables using mains synchronous motors and I was sceptical. https://www.mainsfrequency.com/ has both a live feed and a graph over the last 60 minutes. http://www.mainsfrequency.uk/fm-home has some similar measures. The different measurement methods and weightings for wow and flutter measurements make it difficult to compare directly, but good cassette decks are reported to be of the order of 0.1% peak. That makes me suspect there's a fair chance of a good correlation even with cassette.

 

[Hayabusa]

Hi, I have an audio forensic problem I am trying to resolve.

I have some analog cassette tapes that are fakes that I suspect were digitised, time stretched and then remade to tape. Their length is fixed due to the recording equipment used, so if time stretching could be proven it would identify those responsible in the chain of custody.

If the likely technique could be identified it would greatly assist me in approaching someone with the particular skill.

what suspected amount of stretching are we talking about?

 

[SSS]

If possible change the speed of the cassette recorder in order to get the 50 Hz hum to measured 50 Hz. Although not all recorders allow for modifying the capstan motor. When done, digitize the music.

 

[Tony1]

what suspected amount of stretching are we talking about?

These are 90 min speech tapes (running at half normal speed (15/16 ips) with guessing, say 5 - 10 mins taken out. So not a lot of stretching.

 

[Guermantes]

If it's possible segments have been edited out, then the usual approach is to look for discontinuities in background sounds or even tape noise (some tape transports even make repetitive noise that is picked up by recorder microphones). A good audio editor with spectrogram view is useful for this.

Are these microcassettes or just half-speed compact cassettes? Microcassette recorders have various sources of noise and non-linearities that might leave useful clues to authenticity.

 

[kemmler3D]

These are 90 min speech tapes (running at half normal speed (15/16 ips) with guessing, say 5 - 10 mins taken out. So not a lot of stretching.

So the theory is that some audio was deleted and the remaining audio was stretched to keep the total duration the same?

Makes sense.

I think @Guermantes' approach should be your first stop. Look for any discontinuities in the spectrogram at the time you believe a cut took place. Time stretching would not change this by itself. Matching spectrum at the beginning / end of edits is probably not that hard with modern tools (haven't tried in years) but you would need to know something about audio to realize it was a necessary covering-tracks step.

I would also say 5-10% stretching might be detectable depending on how it was stretched. Speech will tend to sound fairly natural but any extraneous sounds, especially sharp sounds, might end up with unnatural-looking (and sounding?) transients in the stretched version.

 

[Waveform Fidelity]

If possible change the speed of the cassette recorder in order to get the 50 Hz hum to measured 50 Hz. Although not all recorders allow for modifying the capstan motor. When done, digitize the music.

It was commn to have a speed trim control on the board in the early DC servo design, but if I recall correctly later high quality decks used PLL/Quartz Osc control, so it depends on what machine the original recording was made and what machine is now being used for playback.