methods for detection and removal of parasitic frequency modulation in audio recordings czyzewski...
Post on 20-Dec-2015
214 views
TRANSCRIPT
METHODS FORMETHODS FOR DETECTION AND DETECTION AND REMOVAL OF PARASITIC FREQUENCY REMOVAL OF PARASITIC FREQUENCY MODULATION IN AUDIO RECORDINGSMODULATION IN AUDIO RECORDINGS
CZYZEWSKI ACZYZEWSKI A., MAZIEWSKI P., DZIUBINSKI M., KACZMAREK A., ., MAZIEWSKI P., DZIUBINSKI M., KACZMAREK A., KULESZA M., CIARKOWSKI A. & KULESZA M., CIARKOWSKI A. & KOSTEK B.KOSTEK B.
Multimedia Systems DepartmentMultimedia Systems DepartmentGdańsk University of Technology (GUT)Gdańsk University of Technology (GUT)PolandPoland
ScheduleSchedule
Wow & flutter – basic facts and Wow & flutter – basic facts and definitionsdefinitions
Methods of estimating parasitic Methods of estimating parasitic frequency modulation in audiofrequency modulation in audio
Our algorithms and softwareOur algorithms and software Sound examplesSound examples ConclusionsConclusions
Wow & flutter –Wow & flutter – parasiteparasite effectseffects- The wow & flutter The wow & flutter defects can be founddefects can be found in: in:
- RecordingsRecordings, , especiallyespecially archivalarchival on on magnetic and optical sound tapesmagnetic and optical sound tapes
- Those defects can occur during:Those defects can occur during:- ssound recordingound recording- ccopying procedureopying procedure
- The main sources of those defects are:The main sources of those defects are:- irregular velocity of the sound carrier movementirregular velocity of the sound carrier movement- mechanical tape damages mechanical tape damages
- Parasite wowParasite wow (flutter) (flutter) distortion distortionss can be characteri can be characterisseedd as as an undesirable changes of all sound frequency componentsan undesirable changes of all sound frequency components
- Eliminating them should help to understandEliminating them should help to understand the content of some still the content of some still unresolved archival speech recordingsunresolved archival speech recordings
DriftDrift ffmodmod < 0.5Hz < 0.5Hz
WowWow ffmodmod < 0.5 – 6Hz > < 0.5 – 6Hz >
FlutterFlutter ffmodmod < 6Hz – < 6Hz –
100Hz >100Hz >
Frequency-modulation noFrequency-modulation noiise se ffmodmod > 100Hz > 100Hz
Wow & Flutter problem
Standards related with wow & Standards related with wow & flutterflutter
Standards related with wow & Standards related with wow & flutterflutter
Examples of well known standards :Examples of well known standards :
Magnetic Tape Recording and Reproducing (Magnetic Tape Recording and Reproducing (Reel-to-Reel)Reel-to-Reel)NAB Standard NAB Standard 19651965 (National Association of Broadcasters) (National Association of Broadcasters) Messgerate fur Frequenzschwankungen bei Messgerate fur Frequenzschwankungen bei
SchallspeichergeratenSchallspeichergeraten (Measuring Equipment for Frequency (Measuring Equipment for Frequency Variations in Sound Recording Equipment)Variations in Sound Recording Equipment)
DIN DIN 19661966 Measurement of Wow and Flutter in Recording Measurement of Wow and Flutter in Recording
Equipment and in Sound ReproductionEquipment and in Sound ReproductionCCIR Recommendation CCIR Recommendation 19701970
Method for Measurement of Weighted Peak Flutter of Method for Measurement of Weighted Peak Flutter of Sound Recording and Reproducing EquipmentSound Recording and Reproducing Equipment
IEEE 1953 (IEEE 1953 (19711971)) AES 1982 (AES 1982 (20032003))
It is generally accepted that flutter should be less than 0.15% DIN weighted to be inaudible. Measurement methods (2) are described in IEC 386, and a suitable test film is available from the SMPTE (No. P35-FL).
Wow RestorationWow Restoration algorithms algorithms
The defect is difficult The defect is difficult tto o estimateestimate and and to to restorerestore
We We developed several developed several algorithmsalgorithms for for wow restorationwow restoration
Preservation towards storage and access. Standardised Practices for Audiovisual Contents in Europe
FP6-IST-707336
PrestoSpace partnershipPrestoSpace partnership– CoordinatorCoordinator : INA : INA
– Project Steering BoardProject Steering Board : BBC, RAI, INA, ORF, Joanneum Research (Austria), B&G (NL), Sheffield : BBC, RAI, INA, ORF, Joanneum Research (Austria), B&G (NL), Sheffield
University (GB)University (GB)
– PartnersPartners : : CTM Debrie, ACS, Media-Matters, CRCDG, Centrimage, Vectracom, SWR, Front Porch CTM Debrie, ACS, Media-Matters, CRCDG, Centrimage, Vectracom, SWR, Front Porch
Digital, Hi-STor, Houpert Digital Audio, LIMSI-CNRS, Snell&Wilcox, TI Partners, Universities…Digital, Hi-STor, Houpert Digital Audio, LIMSI-CNRS, Snell&Wilcox, TI Partners, Universities…
– Three partners to be identified in second phase of the project Three partners to be identified in second phase of the project
– External User GroupExternal User Group : 30+ considered archives. : 30+ considered archives.
Methods of the restoration of pitch Methods of the restoration of pitch variation defectsvariation defects
based on comparison of the multiple copies of degraded based on comparison of the multiple copies of degraded
audioaudio
Methods of the restoration of pitch Methods of the restoration of pitch variation defectsvariation defects
based on comparison of the multiple copies of degraded based on comparison of the multiple copies of degraded
audioaudio time warping algorithms applied in time or time warping algorithms applied in time or
spectral domain (appropriate algorithm should spectral domain (appropriate algorithm should generate the correct time warping function or generate the correct time warping function or pitch variation function)pitch variation function)..
adaptive filtering (estimation of time delay adaptive filtering (estimation of time delay (offset) between selected copies)(offset) between selected copies)..
correlation methods (finding similarities based correlation methods (finding similarities based on cross-correlation function)on cross-correlation function)..
statistical methods (maximum likelihood statistical methods (maximum likelihood estimation or true Bayesestimation or true Bayesianian estimation – estimation – depending on knowing the priori information)depending on knowing the priori information)..
Methods of the restoration of Methods of the restoration of pitch variation defectspitch variation defects
based on one copy of the archive recordingbased on one copy of the archive recording
Methods of the restoration of Methods of the restoration of pitch variation defectspitch variation defects
based on one copy of the archive recordingbased on one copy of the archive recording speech applications – lowpass filtering speech applications – lowpass filtering
followed by pattern matchingfollowed by pattern matching.. speech - speech - cepstral analysis which reveals cepstral analysis which reveals
information about the fundamental frequencyinformation about the fundamental frequency.. music or speech music or speech - - application application of of methods methods
based on a knowledge of the shape of the based on a knowledge of the shape of the nonlinear function of audio distortionnonlinear function of audio distortion..
PVC PVC = = PPitch itch VVariation ariation CCurveurve
Methods of the restoration of Methods of the restoration of pitch variation defectspitch variation defects
based on one copy of the archive recordingbased on one copy of the archive recording
Methods of the restoration of Methods of the restoration of pitch variation defectspitch variation defects
based on one copy of the archive recordingbased on one copy of the archive recording multiple pitch tracking based on :multiple pitch tracking based on :
– McAulay and Quatieri McAulay and Quatieri representationrepresentation (pitch tracking (pitch tracking through the use of multiple concurrent pitch through the use of multiple concurrent pitch “tracks”)“tracks”)
– Analyzing high-frequency biasAnalyzing high-frequency bias– Analyzing recorded power-line hum (AC power Analyzing recorded power-line hum (AC power
supply)supply) pitch tracking algorithms should generate the pitch tracking algorithms should generate the
correct pitch variation functioncorrect pitch variation function (PVC) (PVC) based on pitch variation based on pitch variation the the restoration of the restoration of the
signal can be done for example by resampling signal can be done for example by resampling it with a truncated sinc functionit with a truncated sinc function..
Our algorithms…Our algorithms…
presto8.exe (7)
CarrierCarrier feature analysis feature analysis TTracking power-line hum frequencyracking power-line hum frequency
Hum is commonly found in archival recordings. Tracking Hum is commonly found in archival recordings. Tracking its frequency allows estimatits frequency allows estimatinging the depth of wow effect. the depth of wow effect. Due to limitations of DFT methods Due to limitations of DFT methods in case of in case of low- low-frequency signals, frequency signals, the the AR pseudospectrumAR pseudospectrum estimation estimation method imethod is used.s used.
Tracking high-frequency biasTracking high-frequency biasAssuming that a high frequency bias was recorded on the Assuming that a high frequency bias was recorded on the
magnetic tape it could serve as an ideal target for magnetic tape it could serve as an ideal target for parasite wow characteristic determining. This idea was parasite wow characteristic determining. This idea was also exploited in the prepared software.also exploited in the prepared software.
Wow Restoration PluginWow Restoration Plugin
WowWow
No No WowWow
R E S T O R E DR E S T O R E D
O R I G I N A LO R I G I N A L
ConclusionsConclusions MQ analysis can be effective in case of parasite frequency MQ analysis can be effective in case of parasite frequency
modulation detectionmodulation detection Formant structure analysis can significantly improve Formant structure analysis can significantly improve
efficiency of the detection processefficiency of the detection process Precise wow tracking can be achieved based on a cepstrally Precise wow tracking can be achieved based on a cepstrally
smoothed spectrum representationsmoothed spectrum representation Recorded power line hum provides a good basis for wow Recorded power line hum provides a good basis for wow
tracking (similarly to high-frequency bias in magnetic tracking (similarly to high-frequency bias in magnetic recordings)recordings)
Interactive procedures (parameters adjustment) leads to Interactive procedures (parameters adjustment) leads to more reliable resultsmore reliable results
It seems that an interactive approach employing an It seems that an interactive approach employing an algorithmic „toolbox” for wow&flutter restoration is the best algorithmic „toolbox” for wow&flutter restoration is the best solution of this problemsolution of this problem
Thank you
for your kind attention