spectrogram.py 文件源码

def work(self, task):
        pts = task

        audio_frame = int(pts * self._api.audio.sample_rate / 1000.0)
        first_sample = (
            audio_frame >> DERIVATION_DISTANCE) << DERIVATION_DISTANCE
        sample_count = 2 << DERIVATION_SIZE

        samples = self._api.audio.get_samples(first_sample, sample_count)
        samples = np.mean(samples, axis=1)
        sample_fmt = self._api.audio.sample_format
        if sample_fmt is None:
            return pts, np.zeros((1 << DERIVATION_SIZE) + 1)
        elif sample_fmt == ffms.FFMS_FMT_S16:
            samples /= 32768.
        elif sample_fmt == ffms.FFMS_FMT_S32:
            samples /= 4294967296.
        elif sample_fmt not in (ffms.FFMS_FMT_FLT, ffms.FFMS_FMT_DBL):
            raise RuntimeError('Unknown sample format: {}'.format(sample_fmt))

        self._input[0:len(samples)] = samples
        out = self._fftw()

        scale_factor = 9 / np.sqrt(1 * (1 << DERIVATION_SIZE))
        out = np.log(
            np.sqrt(
                np.real(out) * np.real(out)
                + np.imag(out) * np.imag(out)
            ) * scale_factor + 1)

        out *= 255
        out = np.clip(out, 0, 255)
        out = np.flip(out, axis=0)
        out = out.astype(dtype=np.uint8)
        return pts, out