SLIDE 1
Characterisation and simulation of telephone channels using the - - PowerPoint PPT Presentation
Characterisation and simulation of telephone channels using the - - PowerPoint PPT Presentation
Characterisation and simulation of telephone channels using the TIMIT and NTIMIT databases Herman Kamper and Thomas Niesler Department of Electrical and Electronic Engineering Stellenbosch University 30 November 2009 Introduction Speech
SLIDE 2
SLIDE 3
Analysis of telephone channels
◮ Used the TIMIT and NTIMIT corpora ◮ Investigated channel (bandlimiting) characteristics ◮ Investigated noise which is added by telephone channel
TIMIT x[n] Telephone channel y[n] NTIMIT
SLIDE 4
Model of the telephone channel
Wideband input x[n] Channel ˆ H(z) u[n]
+ +
Coloured noise Colouring filter ˆ G(z) v[n] White noise σ2
w
w[n] y[n] Bandlimited
- utput
SLIDE 5
Channel analysis
◮ Parametric channel modelling was evaluated (below) ◮ Spectral channel analysis techniques were also evaluated ◮ Used synthetic filters to evaluate the different techniques
TIMIT x[n] Telephone channel NTIMIT
+ −
y[n] Model ˆ H(z) ˆ y[n] e[n]
SLIDE 6
Design of channel model
◮ Analysed the 253 NTIMIT telephone channels ◮ Used a spectral analysis technique ◮ Two possibilities for channel model:
Use filter from channel library Generate random filter based on distributions
1000 2000 3000 4000 5000 6000 7000 8000 −60 −50 −40 −30 −20 −10 10 Frequency (Hz) Amplitude (dB) Average Standard deviation interval
SLIDE 7
Noise analysis I
◮ Used 100 noise segments from arbitrary NTIMIT utterances ◮ Analysed segments to determine spectral characteristics of
additive noise of the NTIMIT telephone channels
◮ Assumed noise segments to be output from LP filters ◮ Designed colouring filter based on the mean LP spectrum
White noise σ2
w
w[n] Colouring filter ˆ G(z) v[n] Coloured noise
SLIDE 8
Noise analysis II
1000 2000 3000 4000 5000 6000 7000 8000 −20 −15 −10 −5 5 10 15 20 25 30 35 Frequency (Hz) Amplitude (dB) Average Median 90% interval
SLIDE 9
Design of noise model
1000 2000 3000 4000 5000 6000 7000 8000 −20 −15 −10 −5 5 10 15 20 25 30 35 Frequency (Hz) Amplitude (dB) Mean LP spectrum Desired amplitude response
SLIDE 10
Implementation in software
Wideband input x[n] Channel ˆ H(z) u[n]
+ +
Coloured noise Colouring filter ˆ G(z) v[n] White noise σ2
w
w[n] y[n] Bandlimited
- utput
SLIDE 11
Evaluation: Single NTIMIT channel I
1000 2000 3000 4000 5000 6000 7000 8000 −100 −90 −80 −70 −60 −50 −40 −30 −20 Frequency (Hz) Power density spectrum (dB) PDS of NTIMIT speech PDS of TIMIT speech
SLIDE 12
Evaluation: Single NTIMIT channel II
1000 2000 3000 4000 5000 6000 7000 8000 −100 −90 −80 −70 −60 −50 −40 −30 −20 Frequency (Hz) Power density spectrum (dB) PDS of NTIMIT speech PDS of y[n] with noise
SLIDE 13
Evaluation: Single NTIMIT channel III
1000 2000 3000 4000 5000 6000 7000 8000 −110 −100 −90 −80 −70 −60 −50 −40 −30 −20 Frequency (Hz) Power density spectrum (dB) PDS of NTIMIT speech PDS of y[n] without noise
SLIDE 14
Evaluation: ASR systems I
TIMIT Software HTK system BPF HTK system NTIMIT Test Test Accuracy Accuracy
SLIDE 15
Evaluation: ASR systems II
Training set Test Set % Accuracy NTIMIT NTIMIT 40.65% TIMIT narrowband NTIMIT 32.56% Filtered TIMIT, 30 dB noise NTIMIT 36.34% Filtered TIMIT, no noise NTIMIT 32.19%
SLIDE 16
Conclusion I
◮ Accuracy obtained using the third system 10.6% lower than
accuracy using the NTIMIT training set
◮ 11.6% increase in accuracy from basic bandpass approach ◮ When no noise is added, performance is not much different
from the TIMIT approach
SLIDE 17