rhdunn / klatt Goto Github PK

According to the Klatt 1990 paper, the F0 flutter delta should be calculated using modulation frequencies of 12.7 Hz, 7.1 Hz and 4.7 Hz The current implementation uses the time_count variable to calculate the amplitudes of these frequencies:

        flc = sin(2*M_PI*12.7*time_count);
	fld = sin(2*M_PI*7.1*time_count);
	fle = sin(2*M_PI*4.7*time_count);

But time_count is incremented once per frame and therefore contains Hz * 100 and not Hz (the unit of time_count is 10ms instead of 1s). So the F0 modulation frequencies are actually too high by a factor of 100.

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

The F0 flutter value, calculated with the 3 sine waves of 12.7, 7.1 and 4.7 Hz, is applied incrementally:

frame->F0hz10 = frame->F0hz10 + (long) delta_f0;

But according to the Klatt 1990 paper, the computed value is already the absolute F0 delta at a specific point in time. It's not a value that has to be integrated over time. The integration has the effect, that F0 wanders around unlimited.

I'm assuming that Klatt never released the source code of this F0 frequency modulation, so we can't refer to the original source code.

The documentation states:

1 = impulse train. 2 = natural simulation. 3 = natural samples

The old source code was:

#define IMPULSIVE        1
#define NATURAL          2
#define SAMPLED          3

The new source code assigns enum values starting with 0:

typedef enum
{
        IMPULSIVE,
        NATURAL,
        SAMPLED,
} klatt_voicing_source_t;

A fix could be either to specify a start value of 1 for the enum:

typedef enum
{
        IMPULSIVE = 1,
        ...

or to subtract 1 from the argument passed on the command line:

            case 'v':
                    globals->glsource = (flag) atoi(optarg) - 1;
                    break;

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

Running example1.par causes a sound spike near the end and example2.par is completely unintelligible when using cascade synthesis. Parallel synthesis works fine.

NOTE: This is also the case with the original 3.03 and 3.04 klatt sources.

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

Example1.par and example2.par generate waveforms are just noise or my operation is wrong?? Is there an example to generate a vowel, word or sentence?

The constant MAX_SAM (Maximum sample rate = 20000) is used to allocate the iwave buffer:

iwave = (int*) malloc(sizeof(int)*MAX_SAM);

But the iwave buffer is used for nspfr entries:

globals->nspfr = (globals->samrate * nmspf_def) / 1000;

for (globals->ns=0; globals->ns < globals->nspfr; globals->ns++)

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.