Using f0

A .f0 file is an ASCII file with two columns containing the time in seconds of the analysed frames and the corresponding fundamental frequencies in Hertz:

time_in_secs_1     f0_value_1
time_in_secs_2     f0_value_2
time_in_secs_3     f0_value_3
    .                  .
    .                  .
    .                  .

Here is an example of an .f0 file :

0.020000        117.934227
0.030000        74.105606
0.040000        98.863930
   .               .
   .               .
   .               .

Here is the "text" conversion of an .f0.sdif file :

SDIF

1NVT
{
StreamID        0;
Date    Wed_Jun__7_15.58.58_2000_;
TableName       GenericBreakPointFunction;
WrittenBy       Pm_Version_1.2.2;
}


SDFC

1FQ0    1       0       0.02
  1FQ0  0x0004  1       1
        117.934

1FQ0    1       0       0.03
  1FQ0  0x0004  1       1
        74.1056

1FQ0    1       0       0.04
  1FQ0  0x0004  1       1
        98.8639
        .
        .
        .
ENDC
ENDF

f0 is available on Unix platforms at Ircam and is called from a command line using the options below :

Necessary parameters

-fefreq : sampling rate in Hz (default : 16000 Hz).

-twidth : FFT width in samples (default : 512).
It is the size of the FFT applied on the signal frame after zero padding. It should be a power of two and greater or equal to the number of samples in the analysed frame of signal.

-istep : analysis step in samples (default : 64).
After each frame analysis, the signal window (or frame) is advanced by this step.

-Nwidth : analysis window width in samples (default : 1024).
This is the size in samples of the signal window (or frame) which is analysed by FFT at each step. In order that spectral peaks appear separated in the FFT analysis, the signal window size should be at least equal to 3 time the inverse of the smaller distance in Hertz between the peaks or partials which should be detected in the analysis.

Facultative parameters

-oiName : input FFT file name.
If this option is not used, the input is the standard input.

-ooName : output file name.
If this option is not used, the output is the standard output.

-fmf0_min
-fMf0_max :
Lower limit and upper limits of the interval within which fundamental frequencies are searched. Defaults are 50 and 200 Hz respectively. It is safer, when possible, to limit the interval [f0_min, f0_max] to one octave. The interval can be adjusted at best after a first f0 detection pass before starting another. In any case, compare the fundamental frequency of your original file (e.g. by hear or by looking at the spacing of the partials on a spectrum as in the image above above) to the -fm and -fM limits and to the result of the analysis, octave errors among other are frequent and often result from wrong settings of f0_min and f0_max.

-fvmf0_min_file
-fvMf0_max_file :

f0_min_file and f0_max_file are files which contain ( respectively ) time varying lower limits and time varying upper limits of the interval within which fundamental frequencies are searched. They are ASCII files with two columns containing the time in seconds and the corresponding limit frequency in Hertz (i.e. what is known as Break Point Function Files or Piece Wise Linear Function):

time_in_secs_1        f0_value_1
time_in_secs_2        f0_value_2
time_in_secs_3        f0_value_3
    .                     .
    .                     .
    .                     .

N.B. time_in_secs_i does not have to be the time of analysised frame i, for each frame, the value of the limit frequency is obtained by linear interpolation.

-Ffreq : The f0_estimation is based on regular spacing of possible harmonic partials up to this frequency. Only partial frequencies below freq are considered ( default : fe/2 ).

-fdfreq : When the spectrum is equal to zero, f0 is set to freq (default : 0).

-bnumber : number of the first fft analysis taken into account (default : 1).

-enumber : number of the last fft analysis into account (default : last one).

-snfloor : This value specifies the noise floor to be used when calculating the fundamental frequency for the soundfile (default : 50dB).

-jsubdivision : histogram bands subdivision (default : 2).

-lpercentage : f0 percentage for the probality law (default : 0.4).

-P : frame times start at i/fe - Nwidth/2fe instead of Nwidth/2fe.

-S : SDIF output of long or short f0 results.

-L : long f0 output file

By default, the format of the fundamental frequency file is as described above (.f0 and .f0.sdif files). The option -L causes the fundamental frequency file format to be extended with more information :

A .f0.long file is an ASCII file with five columns containing the time in seconds of the analysed frames, the corresponding fundamental frequencies, score, real amplitude, confidence good pitch :

time_in_secs_1     f0_value_1  score_1  real_amplitude_1   confidence_1
time_in_secs_2     f0_value_2  score_1  real_amplitude_1   confidence_1 
time_in_secs_3     f0_value_3  score_1  real_amplitude_1   confidence_1 
    .                  .          .            .              .
    .                  .          .            .              .
    .                  .          .            .              .

Here is an example of an .f0.long file :

0.020000  117.934227 72.390953 0.000863 0.534666
0.030000  74.105606 121.991577 0.001941 0.458296
0.040000  98.863930 91.781097 0.001888 0.509323
 .         .         .         .        .
 .         .         .         .        .
 .         .         .         .        .

Here is the "text" conversion of an .f0.long.sdif file :

SDIF

1NVT
{
StreamID        0;
Date    Fri_Jun__9_11.41.21_2000_;
TableName       GenericBreakPointFunction;
WrittenBy       Pm_Version_1.2.2;
}


SDFC

1FQ0    1       1073832416      0.02
  1FQ0  0x0004  1       4
        117.934 72.391  0.000862592     0.534666

1FQ0    1       1073832416      0.03
  1FQ0  0x0004  1       4
        74.1056 121.992 0.00194089      0.458296

1FQ0    1       1073832416      0.04
  1FQ0  0x0004  1       4
        98.8639 91.7811 0.00188833      0.509323
	        .
		.
		.

-H : output whole histogram instead of max value.

Example

f0 -S -fe48000 -t2048 -N256 -i32 -fvmFileMin -fvMFileMax -sn30 < InputFile > OutputFile
(FileMin and FileMax are Min/Max detectable frequency files)