EEL 6586: AUTOMATIC SPEECH PROCESSING Windows Lecture

1
EEL 6586 AUTOMATIC SPEECH PROCESSINGWindows
Lecture

• Mark D. Skowronski
• Computational Neuro-Engineering Lab
• University of Florida
• February 10, 2003

2
No, not MS Windows
3
not those either!
4
Speech windows
Speech is NONSTATIONARY
5
Speech windows
SEVEN

• Assume speech is stationary over short window
  of time.

6
What is a short window of time?

• 10 µs smallest difference detectable by auditory
  system (localization),
• 3 ms shortest phoneme (plosive burst),
• 10 ms glottal pulse period,
• 100 ms average phoneme duration,
• 4 s exhale period during speech.

Short depends on application.
7
Applications using windows

• Automatic speech recognition,
• Speech coding/decoding,
• Speaker identification,
• Text-to-speech synthesis,
• Noise reduction

Typical window (frame) length 20-30 ms Typical
frame rate 100 frames/sec
8
Short-time analysis
s(n) entire speech utterance w(n) window
function x(n) frame of speech
Window function is non-zero for N samples,
n0,,N-1
9
Short-term Fourier Transform
s(m) entire speech utterance w(m) window
function X(n,?) STFT of speech at time n
STFT is a smoothed version of original spectrum.
10
STFT example
s(n) pure sinewave of infinite length w(n)
rectangular window
11
STFT example
W(?)

X(?)

12
Window types

• Rectangular
• Hann (cosine)
• Hamming (raised cosine)
• Blackman
• Kaiser-Bessel

Tradeoff between leakage and blurring
13
Window tradeoff

• Blurring main lobe width A
• Leakage side lobe suppression B

B
A
14
Popular windows
Window Unit BW Sidelobe
Rectangle 1 -13 dB
Hann 2 -31 dB
Hamming 2 -43 dB
Blackman 3 -68 dB
Kaiser-Bessel 4 -91 dB
15
Practical issues

• Rule of thumb
• Time domain, use Rectangle window
• Freq domain, use Hamming window
• Why?

16
Time domain issues

• Correlation in time domain interfered by tapered
  windows

20 ms /eh/, male utterance, pitch measurement
(normalized autocorrelation). First side peak
lower using Hamming window
17
Frequency domain issues
fs12.5 KHz, /eh/, 800 samples, male
speaker. Blurring/Leakage tradeoff evidence