Class thl::dsp::analysis::SpectrumAnalyser#
-
class SpectrumAnalyser : public thl::dsp::fft::StftAnalyser#
Inheritance diagram for thl::dsp::analysis::SpectrumAnalyser:
Collaboration diagram for thl::dsp::analysis::SpectrumAnalyser:
The curve of a spectrum analyser display: an EQ’s analyser, a mastering meter.
Frames come from an StftAnalyser (fed with push(), or drained from a SampleProbe with read_from()). Every frame makes one curve of num_points() values in dB, at frequencies() spaced evenly on a log axis from min_frequency to max_frequency:
The channels are averaged in power (a stereo signal reads as its mean).
Without smoothing, a point is the largest bin between it and its neighbours (a sine keeps its level however many bins fall on one point); where a point lies between two bins, the power between them. With fractional-octave smoothing, a point is the mean power over its band (a sine is spread over it).
Tilt: tilt_db_per_octave added per octave around 1 kHz; 4.5 dB/octave (the default) is how the ear weighs loudness, and shows pink noise about flat.
Speed: a point rises at once and falls by release_db_per_second at most (the display’s ballistics). Freeze: it does not fall at all, so the curve holds the maximum.
Below floor_db everything reads floor_db (the display’s range). Scaled as StftAnalyser: a full-scale sine at a bin’s frequency reads 0 dB, tilt 0.
Threads: everything on one thread, not the audio thread: curve_db() and frequencies() are views of what push(), read_from() and the setters rewrite (set_resolution() reallocates them). A UI on another thread copies the curve on this one. prepare() and set_resolution() allocate; push(), read_from(), the other setters and curve_db() do not.
Public Types
Public Functions
-
bool prepare(double sample_rate, size_t num_channels, const Settings &settings)#
Frames of the resolution’s size, four a frame (75 % overlap), for
num_channelschannels.A maximum frequency above Nyquist is clamped to it. at
sample_rate. False, and unprepared, for settings it cannot use.
-
bool set_resolution(Resolution resolution)#
Re-prepares at another FFT size (allocates); the curve starts over.
-
void set_tilt(float db_per_octave)#
Applies to the curve at once, frozen or not.
-
void set_smoothing(float octaves)#
The curve starts over (a band’s maximum and its mean do not convert).
-
void set_release(float db_per_second)#
-
inline void set_freeze(bool freeze)#
Frozen, the curve only rises; unfrozen it falls again from the maximum it held.
-
void set_floor(float floor_db)#
-
void read_from(thl::dsp::probe::SampleProbe &probe)#
Drains
probe(its one reader) and feeds what it held.After a discontinuity the frames start over (no frame spans the gap); the curve goes on falling from where it was. The probe must have num_channels() channels: otherwise nothing is read (an assert in debug builds).
-
inline std::span<const float> frequencies() const#
The frequency (Hz) of every point.
-
inline std::span<const float> curve_db() const#
The curve in dB, as of the last frame (floor_db() before the first).
Protected Functions
-
virtual void process_spectrum(size_t channel, std::span<const std::complex<float>> bins) override#
One frame of channel
channel:num_bins() bins, scaled as described above.Called for every channel of a frame in turn.
-
struct Settings#
Public Members
-
Resolution m_resolution = Resolution::High#
-
Window m_window = Window::Hann#
BlackmanHarris for a range beyond 90 dB.
-
size_t m_num_points = 256#
-
float m_min_frequency = 20.0f#
-
float m_max_frequency = 20000.0f#
-
float m_tilt_db_per_octave = 4.5f#
-
float m_smoothing_octaves = 0.0f#
0 off, else e.g. 1/3, 1/6, 1/12, 1/24
-
float m_release_db_per_second = 30.0f#
the speed
-
float m_floor_db = -120.0f#
-
bool m_freeze = false#
-
Resolution m_resolution = Resolution::High#