MelFilterBank_Emu <-
function(fs=16000, N = 512, m = 10)
{
# Jonathan Harrington 21.1.22
  # builds a filter-bank of m mel-scaled filters
  # scaled in amplitude between 0 and 1
  # between 0 and fs/2 Hz assuming the FFT was
  # derived from an N = 512 point window
  mat = NULL
  melfreq = seq(0, mel(fs/2), length=m+2)
# Hz centre frequencies
hzfreq = mel(melfreq, inverse=T)
# get spectral components of centre frequencies
k = floor((N+1)*hzfreq/fs) + 1
  
# build the filter bank
for(j in 1:(length(k)-2))
{
amp = c(seq(0, 1, length=k[j+1]-k[j]+1), seq(1, 0, length=k[j+2]-k[j+1]+1)[-1])
freq = seq(k[j], k[j+2], by = 1)
centrefreq = freq[amp==1]
fbank = rep(j, k[j+2]-k[j]+1)
mat$amp = c(mat$amp, amp)
mat$freq = c(mat$freq, freq)
mat$centre = c(mat$centre, rep(centrefreq, length(amp)))
# convert to Hz
mat$fbank = c(mat$fbank, fbank)
}
# convert spectral component to Hz
mat$Hz = fs/N * (mat$freq-1)
mat$Hzcen = fs/N * (mat$centre-1)
mat$Melcen = mel(mat$Hzcen)
# output as data-frame
d = data.frame(A = mat$amp, freq = mat$Hz, Hzcen = mat$Hzcen,
                    Melcen = mat$Melcen,
                    k = mat$freq, fbank = mat$fbank)
as_tibble(d)
}