fbank

Functions

int main(int argc, char *argv[])

fbank [ option ] [ infile ]

-n int
- number of channels \((1 \le C)\)
-l int
- FFT length \((2 \le N)\)
-s double
- sampling rate in kHz \((0 < F_s)\)
-L double
- lowest frequency in Hz \((0 \le F_l < F_h)\)
-H double
- highest frequency in Hz \((F_l < F_h \le 500F_s)\)
-q int
- input format
  - 0 amplitude spectrum in dB
  - 1 log amplitude spectrum
  - 2 amplitude spectrum
  - 3 power spectrum
  - 4 windowed waveform
-o int
- output format
  - 0 fbank
  - 1 fbank and energy
-e double
- floor of raw filter-bank output \((0 < \epsilon)\)
infile str
- double-type windowed sequence or spectrum
stdout
- double-type mel-filter-bank output

The below example extracts the 20-channel mel-filter-bank outputs from a Hamming windowed signal.

frame -l 400 -p 160 < data.d | window -l 400 -L 512 -w 1 | \
   fbank -l 512 -n 20 > data.fbank

Parameters:

argc – [in] Number of arguments.
argv – [in] Argument vector.

Returns:

0 on success, 1 on failure.

See also

mfcc

class MelFilterBankAnalysis

Perform mel-filter-bank analysis.

The input is the half part of power spectrum:

\[ \begin{array}{cccc} |X(0)|^2, & |X(1)|^2, & \ldots, & |X(N/2)|^2, \end{array} \]

where \(N\) is the FFT length. The outputs are the \(C\)-channel mel-filter-bank outputs

\[ \begin{array}{cccc} F(1), & F(2), & \ldots, & F(C) \end{array} \]

and the log-signal energy \(E\).

The implementation is based on HTK. The only difference from the implementation is the constant of mel-scale formula:

\[ m = 1127.01048 \log \left( 1 + \frac{f}{700} \right), \]

where HTK use \(1127\) instead of \(1127.01048\).

[1] S. Young et al., “The HTK book,” Cambridge University Engineering Department, 2006.

Public Functions

MelFilterBankAnalysis(int fft_length, int num_channel, double sampling_rate, double lowest_frequency, double highest_frequency, double floor, bool use_power)

Parameters:

fft_length – [in] Number of FFT bins, \(N\).
num_channel – [in] Number of channels, \(C\).
sampling_rate – [in] Sampling rate in Hz.
lowest_frequency – [in] Lowest frequency in Hz.
highest_frequency – [in] Highest frequency in Hz.
floor – [in] Floor value of raw filter-bank output.
use_power – [in] If true, use power spectrum instead of amplitude one.

inline int GetFftLength() const

Returns:: FFT size.

inline int GetNumChannel() const

Returns:: Number of channels.

inline double GetFloor() const

Returns:: Floor value.

inline bool IsPowerUsed() const

Returns:: Whether to use power spectrum.

inline bool IsValid() const

Returns:: True if this object is valid.

bool GetCenterFrequencies(std::vector<double> *center_frequencies) const

Returns:: Center frequencies in Hz.

bool Run(const std::vector<double> &power_spectrum, std::vector<double> *filter_bank_output, double *energy) const

Parameters:

power_spectrum – [in] \((N/2+1)\)-length power spectrum.
filter_bank_output – [out] \(C\)-channel filter-bank outputs.
energy – [out] Signal energy \(E\) (optional).

Returns:

True on success, false on failure.