centro ricerche e innovazione tecnologica presto - preservation technologies for european broadcast...
TRANSCRIPT
Centro Ricerche e Innovazione Tecnologica
Presto - Preservation Technologies forEuropean Broadcast Archives
Work Package 4
Status Report
Centro Ricerche e Innovazione Tecnologica
Test bed architecture
A/D Converter 24 bit@48 KHz
Player control interfaceTape/Vynil
Player
BWF
BWF
BWF
XML
XML
XML (House keeping data)
Mp3/BWF
Quality Control(acquisition monitor)
Acquisition&
Player Control
BWF
MP3 encoder
Quality Control(process monitor)
LosslessCompression
Key link 4.3Key link 4.4
Key link 4.1
Key link 4.2
Centro Ricerche e Innovazione Tecnologica
WP4 workplan
Centro Ricerche e Innovazione Tecnologica
Presto - Preservation Technologies forEuropean Broadcast Archives
key link 4.1:
QUALITY MONITOR
Centro Ricerche e Innovazione Tecnologica
Audio Quality Monitor
Algorithms developed for:
computations of signal features
silence detection
bandwidth computation
click detection Algorithm under construction:
channel phase correlation
Centro Ricerche e Innovazione Tecnologica
Signal Features
power: estimated through the mean square amplitude
peak level: maximum sample value
dynamic: difference between maximum and minimum
sample values
presence of DC: estimated through the mean of sample
values
Centro Ricerche e Innovazione Tecnologica
Silence Detection
Robust detection of silence can be based on the comparison of values of the following parameters with properly set thresholds:
absolute energy
ratio between local and average energy
sample variance
sample maximum
For light computation, only sample variance is used.
Centro Ricerche e Innovazione Tecnologica
Bandwidth Computation
The power spectral density of a function s(t), sampled at N
points to produce FFT values c0-cN-1, can be estimated
by the sum squared amplitude (SSA):
Let k the index such that:
that is, k is the index that split the total power in .
Then, the bandwidth can be estimated as (F=sample freq.):
1 ,
Centro Ricerche e Innovazione Tecnologica
Click DetectionThe audio signal is modeled with autoregressive (AR) or all pole model:
According to AR model, the observed data x[t], corrupted
by clicks, is filtered using prediction error filter:
In this way the detection signal is:
The coefficients ai are estimated by the Maximum Entropy Method.
Centro Ricerche e Innovazione Tecnologica
Integration into Elettra System
process monitor (on-line): all the processing functions are integrated through a library into the Elettra system:
input: buffer of data samples (size configurable)
output: log file + interface with visual tools
acquisition monitor (off-line): it is an executable program:
input: BWF file (audio 48KHz, 24bit, stereo)
output: XML file (quality analysis)
Centro Ricerche e Innovazione Tecnologica
Exploitation (1)
As stated in the Key Links System Specification Document, ITC-irst will release two sets of libraries:
Video Quality Control Audio Quality Control
compiled for Linux OS, and documentation.
The packages can be exploited in the following ways:
within and beyond Presto, by all the project partners for demos of Presto results
beyond Presto, even for commercial purposes after an agreement with ITC-irst
Centro Ricerche e Innovazione Tecnologica
Exploitation (2)
Usually, ITC-irst does not release source codes
exploited for commercial purposes, neither
guarantees heavy support.
More relaxed conditions are applied if exploitation
regards demos and research. However, we decide
case by case, and all the aspects of the
agreement are defined during the negotiation.
Centro Ricerche e Innovazione Tecnologica
Presto - Preservation Technologies forEuropean Broadcast Archives
key link 4.3:
PLAYBACK DEVICE IMPROVEMENTS
Centro Ricerche e Innovazione Tecnologica
Turntable improvement
Double arm for 78 RPM Two different pickups can be used at the same time:
E.g. conical vs. elliptical stylus or .0040” vs. .0028” stylusBoth output can be captured and compared in the digital
domain with the aid of Quality Control Commercial solutions are disappearing from the
marketStart/stop automation for 33 RPM
Automatic synchronisation between reproduction and capturing equipment under software control
No commercial solutions available
Centro Ricerche e Innovazione Tecnologica
Activities carried out so far
Interface control document available Command set: start, stop, status Physical layer: RS232
Basic turntable components selected Plate, motor, electronics: Technics 1200 Tonearm: SME 309/312. Up to 2 arms per turntable
First prototype mechanical design completed Turntable base and arm fixing block Arm lift device Stop detector
Physical realization is under way
Centro Ricerche e Innovazione Tecnologica
A/D conversion technology selection (1)
Stagetec Reference Master 24 bit quad converter tested against Apogee PSX-100 24 bit Sigma/Delta conversion unit
Extensive set of objective measurements have been carried out THD+N curves are significantly different Stagetec shows extended dynamic range, close to
theoretical values
Centro Ricerche e Innovazione Tecnologica
A/D conversion technology selection (2)
Subjective tests (expert panel listening) Test materials:
Live recording of the RAI orchestraRecording of vynil records (33 and 78 RPM) at nominal levelRecording of vynil records at low level (30 dB below nominal)
Test set upThe listening was performed with both loudspeakers (Genelec
S30c and 1038A and Dynaudio) and with headphones (Stax Lambda Pro with diffuse field eq) using Apogee PSX-100 24 bit D/A converter for all the materials
Low level materials were played after rescaling at nominal level
5 expert listeners took part in the evaluation
Centro Ricerche e Innovazione Tecnologica
Preliminary resultsThe listeners were hardly able to detect any difference on
nominal level recorded materials– Altought this could be due to the insufficient resolution of the
D/A converter, it seems that THD+N values lower then -100 dB are scarcely perceptible during normal listening
The extended dynamic range of Stagetec was clearly perceivable on low level recorded materials
Further tests will be performed by applying restoration techniques (denoise, declick) to vynil recorded materials
A/D conversion technology selection (3)
Centro Ricerche e Innovazione Tecnologica
Presto - Preservation Technologies forEuropean Broadcast Archives
key link 4.4:
LOSSLESS COMPRESSION
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
key link: Lossless Compression Tecnology (LCT) With Lossless Compression Tecnology is possible to
reduce the storage necessary for the digitized material
24bit - 48kHzlinear
24bit - 48kHzlinear
24bit - 48kHzcompressed
24bit - 48kHzcompressed
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Any Lossless Compression Module (LCM) can be considered as a trasparent layer for the applications accessing the archive storage area
BWF format
BWF format
in
out
LCM codec Digital archive
24bit/48kHzcompressed24bit/48kHzcompressed
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Software requirement: Integration in an acquisition automatic process
(command line version, etc..) algorithm robustness Public source code / Open Project Platform independence Encoding/decoding faster than real time Partial file decoding
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Market survey - Main software features
Product Version Source available OS supportFLAC 1.0 yes anyLPAC 1.31 no Win/Linux
Monkey's Audio 3.92b no windows onlyPkZip 4.00 no anyRKAU 1.07 no windows only
WavPack 3.9 no windows only
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Test Material
Source: 33 RPM vinyl Status: sufficient/good Material: 7 item selection from classic, jazz, rock&pop
records Total duration: about 107 minutes
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Test Material
Two level test: step 1: broadcast quality @ 24 bit / 48kHzstep 2: CD quality @ 16 / bit 48kHz
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Compression Ratio
0,0
10,0
20,030,0
40,0
50,0
60,0
70,080,0
90,0
100,0
24 bit 16bit
%
FLAC
LPAC
Monkey's Audio
PkZip
RKAU
WavPack
original
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Compression Time Ratio (n:1)
0
5
10
15
20
25
30
35
40
24 bit 16bit
n:1
FLAC
LPAC
Monkey's Audio
PkZip
RKAU
WavPack
n:1 means n times faster than realtime
Centro Ricerche e Innovazione Tecnologica
Presto ProjectWP4 - AUDIO DIGITISATION
Hardware Platform
Pentium III @ 1 GHz 256 Mbyte RAM (DDR 266 MHz) OS Windows 2000