Waterfall display of the Intelligent Speech Analyser™

Home Signal displays 1985- Multi-envelope displays 1998- FFT displays 1985- Cepstrum displays 1983- LPC displays 1992- Auditory displays 1983- Loudness displays 1995- Loudness curves 1998- Timbre spectrum 1983- LTAS displays 1982- Harmonic FFT displays 1989- F0 displays 1983- F0 histograms in semitone scale 1989- Jitter/shimmer 1990- Computer Voice Fields 1983- FFT spectrogram displays 1985- LPC spectrogram displays Auditory spectrogram displays 1985- Bark, ERB, mel, semitone and Hz scales formant charts 1988- Synte 2 demo video speech on 8.8.1977 analyzed 2018- Formant charts of world languages in Bark scale 2018- Other links

Works Works before ISA

Spectrum series

FFT-spektrisarja dB/Hz-asteikolla.

|X(k)|_dB (k=0...N/2-1) = 20log₁₀|SUM(w(n)x(n)e^-jk(2π/N)n)|, n=0...N-1. Signal x(n) = x_a(nT), n=0...N-1, T is the sample period, x_a= analog signal. If x_a is the voltage signal of the microphone, then x_a(nT) = kP_a(nT), where P_a is the pressure and k is the constant factor.

The computation uses the decimation-in-time (DIT) computation method. Over the years, I have coded as a DSP man FFT spectrum of the Intelligent Speech Analyser™ (ISA) program
(1) for the Texas 16-bit TMS320 signal processor family in machine language,
(2) for Motorola's 16 and 32-bit M68000 microprocessor families in machine language and C language,
(3) for IBM 600 Series 32-bit PowerPC Microprocessor Family in machine language and C language,
(4) for Intel 32-bit and 64-bit microprocessor families in C++ language.

Signal windowing is done with a Hamming window, a rectangular window or a Blackman-Harris window with more than 90 dB.

Analysis pictures I have coded from the very beginning in Neon object-oriented programming language.

Hamming-window w(n) = 0.54-0.46cos(2πn/(N-1)).

Blackman-Harris-window w(n) = 0.35875-0.48829cos(2πn/(N-1))+0.14128cos(4πn/(N-1))-0.01168 cos(6πn/(N-1)).

Hz is converted to Bark by the formula x/Bark=7ln[f/650+√(1+(f/650)²)] and Barks are converted to Hz by the formula f/Hz=650sinh(x/7).

Hz is converted to ERBs by the formula x/ERB=21,3log₁₀(1+f/228,7) and ERBs are converted to Hz by the formula f/Hz=228,7(10^x/21,3-1).

Hz is converted to mel by the formula x/mel=2595log₁₀(1+f/700) and mels are converted to Hz by the formula f/Hz=700(10^x/2595-1).

The above Bark scale conversion formulas have been in my use since 1983.LPC-spektrisarjat dB/Hz-asteikolla.

The calculation uses the PARCOR method. Over the years, I have coded as a DSP man PARCOR method of the Intelligent Speech Analyser™ (ISA) program
(1) for the Texas 16-bit TMS320 signal processor family in machine language,
(2) for Motorola's 16 and 32-bit M68000 microprocessor families in machine language and C language,
(3) for IBM 600 Series 32-bit PowerPC Microprocessor Family in machine language and C language,
(4) for Intel 32-bit and 64-bit microprocessor families in C++ language.

Signal windowing is done with a Hamming window, a rectangular window or a Blackman-Harris window with more than 90 dB.

Auditoriset spektrisarjat dB/Bark-asteikolla.

Käsitteet "auditorinen" (auditory) ja "auditiivinen" (auditive) ovat eri käsitteitä. ISA:ssa käytetään nimenomaan käsitettä "auditorinen" (auditory).

40 vuoden ajan Bark-asteikko ja auditorinen suodinpankkimalli ovat olleet minulla rutiinikäytössä Otaniemen työvuodestani 1983 lähtien. Kehitin tällöin auditorista suodinpankkimallia yhdessä professori Matti Karjalaisen kanssa. Työskentelin vuosikymmeniä 1970-luvulta saakka Karjalaisen kanssa.
Over the years, I have coded as a DSP man auditory filter bank model completely from zero
(1) for "PDP-11/34 Floating Point Systems FPS 100 Vector Processor" in Fortran,
(2) for the Texas 16-bit TMS320 signal processor family in machine language,
(3) for Motorola's 16 and 32-bit M68000 microprocessor families in machine language and C language,
(4) for IBM 600 Series 32-bit PowerPC Microprocessor Family in machine language and C language,
(5) for Intel 32-bit and 64-bit microprocessor families in C++ language.
Hz is converted to Bark by the formula x/Bark=7ln[f/650+√(1+(f/650)²)] and Barks are converted to Hz by the formula f/Hz=650sinh(x/7).

The above Bark scale conversion formulas have been in my use since 1983.

Kepstrisarjat lin/Hz-asteikolla.

|X(k)|_dB (k=0...N/2-1) = 20log₁₀|SUM(w(n)x(n)e^-jk(2π/N)n)|, n=0...N-1. |Y(k)| (k=0...N/4-1) = |SUM(|X(n)|_dBe^-jk(2π/N)n)|, n=0...N/2-1. Signal x(n) = x_a(nT), n=0...N-1, T is the sample period, x_a= analog signal. If x_a is the voltage signal of the microphone, then x_a(nT) = kP_a(nT), where P_a is the pressure and k is the constant factor.
The computation uses the decimation-in-time (DIT) computation method. Over the years, I have coded as a DSP man Ceptrum of the Intelligent Speech Analyser™ (ISA) program
(1) for the Texas 16-bit TMS320 signal processor family in machine language,
(2) for Motorola's 16 and 32-bit M68000 microprocessor families in machine language and C language,
(3) for IBM 600 Series 32-bit PowerPC Microprocessor Family in machine language and C language,
(4) for Intel 32-bit and 64-bit microprocessor families in C++ language.

Signal windowing is done with a Hamming window, a rectangular window or a Blackman-Harris window with more than 90 dB.

Hamming-window w(n) = 0.54-0.46cos(2πn/(N-1)).

Blackman-Harris-window w(n) = 0.35875-0.48829cos(2πn/(N-1))+0.14128cos(4πn/(N-1))-0.01168 cos(6πn/(N-1)).

|X(k)|dB (k=0...N/2-1) = 20log10|SUM(w(n)x(n)e-jk(2π/N)n)|, n=0...N-1. Signal x(n) = xa(nT), n=0...N-1, T is the sample period, xa= analog signal. If xa is the voltage signal of the microphone, then xa(nT) = kPa(nT), where Pa is the pressure and k is the constant factor.

|X(k)|_dB (k=0...N/2-1) = 20log₁₀|SUM(w(n)x(n)e^-jk(2π/N)n)|, n=0...N-1. Signal x(n) = x_a(nT), n=0...N-1, T is the sample period, x_a= analog signal. If x_a is the voltage signal of the microphone, then x_a(nT) = kP_a(nT), where P_a is the pressure and k is the constant factor.