Missing data speech recognition in rever- berant acoustic conditions - - PowerPoint PPT Presentation

Missing data speech recognition in rever- berant acoustic conditions Kalle, Guy and Jon

ONE SIX FOUR... !

• Mr. M. D. Dummy

Content 1.Introduction to reverberation 2.Speech modulation frequencies 3.Reverberation masking model 4.Test conditions 5.Results 6.Discussion

−15 −10 −5 5 10 15 20 25 30 −15 −10 −5 5 10 15 20 25 30

• 1. Introduction to reverberation
• Fig1. Image expansion. Direct sound.

−200 −150 −100 −50 50 100 150 200 −150 −100 −50 50 100 150

• 1. Introduction to reverberation
• Fig2. Image expansion. 3rd order reflections.

1000 2000 3000 4000 5000 6000 7000 −0.8 −0.6 −0.4 −0.2 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 x 10

4

−120 −100 −80 −60 −40 −20

• 1. Introduction to reverberation
• Fig3. Room impulse responses. Left: linear
• amplitude. Right: log-amplitude.

10 20 30 40 50 60 70 80 5 10 15 20 25 30 20 40 60 80 100 120 140 160 5 10 15 20 25 30

• 2. Speech modulation frequencies
• Fig4. Ratemap for

utterance 5527.

• Fig5. FFT of the

ratemap of Fig3.

10 20 30 40 50 −10 10 20 Magn (dB)

• Freq. (Hz)
• 3. Reverberation masking model

auditory filterbank leaky integrator missing data speech recogniser cube root g rate map threshold downsampling by masking filter compression mask

• Fig6. Diagram of the model.

0.5 1 1.5 2 2.5 0.5 2 6.4 0.5 1 1.5 2 2.5 0.5 2 6.4 0.5 1 1.5 2 2.5 0.5 2 6.4 0.5 1 1.5 2 2.5 0.5 2 6.4

• 3. Reverberation masking model
• Fig7. Rate maps; Top: clean.

Bottom: reverberated

• Fig8. Masks; Top: a-priori.

Bottom: reverb. masking

• 4. Test conditions

Room 1: 15 x 13 x 6.5 m3, T60=0.7 sec., D/R=0 & -10 dB Room 2: 25 x 20 x 6 m3, T60=1.7 sec., D/R=0 & -10 dB Room 3: 55 x 35 x 14 m3, T60=2.7 sec., D/R=0 & -10 dB

• 5. Results

Recognition Technique Clean T60=0.7s D/R= 0dB T60=0.7s D/R=

• 10db

T60=1.7s D/R=0dB T60=1.7s D/R=

• 10dB

T60=2.7s D/R= 0dB T60=2.7s D/R=- 10dB Unity mask 98.26 62.48 46.39 42.82 29.24 34.90 24.28 MFCC 99.65 60.40 47.08 47.35 34.46 40.73 28.55 Reverb. masking 90.33 85.03 82.25 71.11 66.05 45.52 a priori mask 95.82 93.73 92.42 89.12 90.60 87.99

T

• 6. Discussion
• Advantage over robust feature tech-

niques: + Mask estimation can be changed on the fly when condi-

tions change -> no-retraining required when the rule is changed

+ Better performance ???

• Currently all thresholds are hand tuned,

an adaptive system is under work