DAT-error statistics - Brecht Declercq

Post on 26-Jun-2015

397 views 0 download

Tags:

description

As presented on the IASA 2011 World Conference in Frankfurt a.M.

Transcript of DAT-error statistics - Brecht Declercq

DAT-error statistics

Brecht Declercq (VRT) – IASA Conference 2011 (Frankfurt a.M.) 08.09.2011

Typology - risk-analysis - correlationsas registered in VRT’s DivA project

VRT Diva-project

VRT public broadcaster of the Flemish Community in Belgium

VRT-archives 280 000h non file based film & video141 000h non file based audio

DivA-project Digital VRT-archives2008 – 2012€ 2000 000 / year

Digitising goal 13000h audio, 13000h film & video

VERDI 2012 - 2017

DivADIGITIZE & LEARN

DAT-collection @ VRT-archives

Concerts and radio-programmes:

music: # 12000 14000 h radio programmes: # 33000 58000 hTOTAL: # 45000 72000 h

Ingested up till now:

music: # 3500 4000hradio programmes: # 4200 8000h

SHOULD I AND CAN IPRIORITIZEDAT-INGEST BECAUSE OF THE RISK OF DETERRIORATION?AND IF SO:WHAT’S THE CRITERION?AGE?BRAND?

TYPOLOGYWe’ve seen / heard:

- Clicks- Complete signal drop-out- Left-channel, right channel or both

= BLER-errors (Block Error Rate)

- Broken tape

METHODControl during ingest:- Ingest with 6x SONY 7040

- corrupt tapes: 1x rerun with SONY R500- tape problems during ingest: blockage, broken tape- quick visual scan of waveform: major drops

Control during annotation:- Full listening by annotators: check for audible errors- Logging and documentation of errors in report files:

- what kind?- duration?- point in the audiofile

Why not with automatic signal analysis?

Overload of data Difficult to interpret No 1 on 1 relation with audible errors:

- false positives- false negatives- difficult to calibrate

But certainly not useless!

RESEARCH-SAMPLENUMBER OF DATs 1000: daily Mo-Fri programmeRECORD DATE 08/01/1992 – 02/05/1996

= as close as we can get to production date= core period of DAT-usage era at VRT

TRANSFER TO FILE 24/08/2009 – 30/11/2009ANNOTATION & ERROR LOGGING 18/11/2009 – 01/03/2011DURATION 1000 x 120 minBRANDS SONY: 327

AMPEX: 134BASF: 549

SONYAMPEX BASF

CONCLUSIONS:AGE:DATs which contained at least one audible error

RECORD YEAR DATs AFFECTED TOTAL # OF DATs PCT

1992 25 218 11,4

1993 24 224 10,7

1994 17 256 6,6

1995 14 216 6

1996 1 86 1,1

TOTAL 81 1000 8,1

CONCLUSIONS:AGE:Total duration of affected audio within the DAT

RECORD YEAR 1 SEC 1->10 SEC

10 SEC -> 1 MINUTE

1 MINUTE -> TOTAL DAT TOTAL DAT

# % # % # % # % # %

1992 6 2,7 15 6,8 2 0,9 1 0,5 1 0,5

1993 6 2,6 7 3,1 4 1,7 2 0,9 5 2,2

1994 7 2,7 7 2,7 3 1,1 0 0 0 0

1995 4 1,9 4 1,9 3 1,3 1 0,5 2 0,9

1996 1 1,1 0 0 0 0 0 0 0 0

TOTAL 24 2,4 33 3,3 12 1,2 4 0,4 8 0,8

CONCLUSIONS:AGE:Total duration of affected audio within the DAT

1211 minutes affected out of 120 000 minutes of DAT : 1% of total audio was affected, only 120 minutes (1 DAT, 0,1 %) could not be recuperated because of deterrioration.

CONCLUSIONS:BRAND:DATs which contained at least one audible error

BRANDDATs

AFFECTED PCT

SONY 32 9,7

AMPEX 17 12,6

BASF 31 5,6

MINUTE 1 2 -> 5 MINUTES 6 -> 114 115 -> 119 1201992 6 4 9 2 11993 8 14 11 8 61994 2 1 12 1 01995 4 3 7 1 11996 1 0 0 0 0

SONY 9 7 15 6 2

AMPEX 7 8 9 5 5BASF 5 4 24 1 1

CONCLUSIONS:AGE AND BRAND:At which point in the audiofile do errors occur?

CONCLUSIONS:Should I prioritize with age or brand as a criterion?

No clear indication that you should (in my figures). Slight correlations with age (older = worse) and brand (Ampex) need confirmation from more extensive research.

Is the choice completely free then? Traumas seem to play a role.

- recording conditions (e.g. maintenance of devices) - conservation circumstances (temperature, humidity, …)

Try to know the history of the collection if possible. Priority to traumatised tapes

Scarcity of devices (and competences in maintenance and

repair) is the bigger problem. What about DDS?

CONCLUSIONS:NUANCES:

Not always easy to know if an artefact was created- During recording at the time.- As a result of deterrioration.- During the ingest.

This is only one study, many more should be done. Not all brands, not the whole DAT era. Deeper research is complicated (e.g. certain periods of certain brands?) Specific circumstances <-> every collection is different

- recording with good machines- good conservation

ADVICE:

Follow the classic advices for ingest and conservation. Digitizing is a learning process:

- document your process and results carefully- adapt your prioritization in case of clear relations- publish your results

CONCLUSIONS:

I’ve been bombing you with questions!How about a counter attack?

brecht.declercq@vrt.be

http://be.linkedin.com/in/brechtdeclercq

#brecht_declercq