HexoDSP/tests/common/mod.rs
2021-08-04 03:58:43 +02:00

631 lines
16 KiB
Rust

// Copyright (c) 2021 Weird Constructor <weirdconstructor@gmail.com>
// This file is a part of HexoDSP. Released under GPL-3.0-or-later.
// See README.md and COPYING for details.
pub use hexodsp::matrix::*;
pub use hexodsp::nodes::new_node_engine;
pub use hexodsp::dsp::*;
pub use hexodsp::NodeExecutor;
use hound;
pub const SAMPLE_RATE : f32 = 44100.0;
#[allow(dead_code)]
pub const SAMPLE_RATE_US : usize = 44100;
#[macro_export]
macro_rules! assert_float_eq {
($a:expr, $b:expr) => {
if ($a - $b).abs() > 0.0001 {
panic!(r#"assertion failed: `(left == right)`
left: `{:?}`,
right: `{:?}`"#, $a, $b)
}
}
}
#[macro_export]
macro_rules! assert_fpair_eq {
($a:expr, $b:expr) => {
if ($a.0 - $b.0).abs() > 0.0001 {
panic!(r#"assertion failed: `(left.0 == right.0)`
left: `{:?}`,
right: `{:?}`"#, $a.0, $b.0)
}
if ($a.1 - $b.1).abs() > 0.0001 {
panic!(r#"assertion failed: `(left.1 == right.1)`
left: `{:?}`,
right: `{:?}`"#, $a.1, $b.1)
}
}
}
#[macro_export]
macro_rules! assert_f3tupl_eq {
($a:expr, $b:expr) => {
if ($a.0 - $b.0).abs() > 0.0001 {
panic!(r#"assertion failed: `(left.0 == right.0)`
left.0: `{:?}`,
right.0: `{:?}`"#, $a.0, $b.0)
}
if ($a.1 - $b.1).abs() > 0.0001 {
panic!(r#"assertion failed: `(left.1 == right.1)`
left.1: `{:?}`,
right.1: `{:?}`"#, $a.1, $b.1)
}
if ($a.2 - $b.2).abs() > 0.0001 {
panic!(r#"assertion failed: `(left.2 == right.2)`
left.2: `{:?}`,
right.2: `{:?}`"#, $a.2, $b.2)
}
}
}
#[macro_export]
macro_rules! assert_vec_feq {
($vec:expr, $cmp_vec:expr) => {
let cmp_vec = $cmp_vec;
let res : Vec<f32> = $vec.iter().copied().collect();
for (i, (s, scmp)) in res.iter().zip(cmp_vec.iter()).enumerate() {
if (s - scmp).abs() > 0.0001 {
panic!(r#"
table_left: {:?}
table_right: {:?}
assertion failed: `(left[{}] == right[{}])`
left: `{:?}`,
right: `{:?}`"#, &res[i..], &(cmp_vec[i..]), i, i, s, scmp)
}
}
}
}
#[macro_export]
macro_rules! assert_decimated_feq {
($vec:expr, $decimate:expr, $cmp_vec:expr) => {
let cmp_vec = $cmp_vec;
let res : Vec<f32> = $vec.iter().step_by($decimate).copied().collect();
for (i, (s, scmp)) in res.iter().zip(cmp_vec.iter()).enumerate() {
if (s - scmp).abs() > 0.0001 {
panic!(r#"
table_left: {:?}
table_right: {:?}
assertion failed: `(left[{}] == right[{}])`
left: `{:?}`,
right: `{:?}`"#, &res[i..], &(cmp_vec[i..]), i, i, s, scmp)
}
}
}
}
#[macro_export]
macro_rules! assert_slope_feq {
($vec:expr, $cmp_vec:expr) => {
let cmp_vec = $cmp_vec;
let mut res : Vec<f32> = vec![];
let mut prev = 0.0;
for (i, s) in $vec.iter().enumerate() {
let delta = *s - prev;
if i > 0 {
res.push(delta);
}
prev = *s;
}
let res : Vec<f32> = res.iter().copied().collect();
for (i, (s, scmp)) in res.iter().zip(cmp_vec.iter()).enumerate() {
if (s - scmp).abs() > 0.0001 {
panic!(r#"
table_left: {:?}
table_right: {:?}
assertion failed: `(left[{}] == right[{}])`
left: `{:?}`,
right: `{:?}`"#, &res[i..], &(cmp_vec[i..]), i, i, s, scmp)
}
}
}
}
#[macro_export]
macro_rules! assert_decimated_slope_feq {
($vec:expr, $decimate:expr, $cmp_vec:expr) => {
let cmp_vec = $cmp_vec;
let mut res : Vec<f32> = vec![];
let mut prev = 0.0;
for (i, s) in $vec.iter().enumerate() {
let delta = *s - prev;
if i > 0 {
res.push(delta);
}
prev = *s;
}
let res : Vec<f32> = res.iter().step_by($decimate).copied().collect();
for (i, (s, scmp)) in res.iter().zip(cmp_vec.iter()).enumerate() {
if (s - scmp).abs() > 0.0001 {
panic!(r#"
table_left: {:?}
table_right: {:?}
assertion failed: `(left[{}] == right[{}])`
left: `{:?}`,
right: `{:?}`"#, &res[i..], &(cmp_vec[i..]), i, i, s, scmp)
}
}
}
}
#[macro_export]
macro_rules! assert_decimated_slope_feq_fine {
($vec:expr, $decimate:expr, $cmp_vec:expr) => {
let cmp_vec = $cmp_vec;
let mut res : Vec<f32> = vec![];
let mut prev = 0.0;
for (i, s) in $vec.iter().enumerate() {
let delta = *s - prev;
if i > 0 {
res.push(delta);
}
prev = *s;
}
let res : Vec<f32> = res.iter().step_by($decimate).copied().collect();
for (i, (s, scmp)) in res.iter().zip(cmp_vec.iter()).enumerate() {
if (s - scmp).abs() > 0.0000001 {
panic!(r#"
table_left: {:?}
table_right: {:?}
assertion failed: `(left[{}] == right[{}])`
left: `{:?}`,
right: `{:?}`"#, &res[i..], &(cmp_vec[i..]), i, i, s, scmp)
}
}
}
}
#[allow(dead_code)]
pub fn collect_signal_changes(inp: &[f32], thres: i64) -> Vec<(usize, i64)> {
let mut idxs = vec![];
let mut last_sig = 0.0;
for i in 0..inp.len() {
if (inp[i] - last_sig).abs() > 0.1 {
idxs.push((i, (inp[i] * 100.0).floor() as i64));
last_sig = inp[i];
}
}
let mut idxs_big = vec![];
for v in idxs.iter() {
if v.1.abs() > thres {
idxs_big.push(*v);
}
}
return idxs_big;
}
#[macro_export]
macro_rules! assert_rmsmima {
($rms:expr, $b:expr) => {
assert_f3tupl_eq!($rms, $b);
}
}
#[macro_export]
macro_rules! assert_minmax_of_rms {
($rms:expr, $b:expr) => {
let (_, min, max) = $rms;
assert_fpair_eq!((min, max), $b);
}
}
#[allow(unused)]
pub fn pset_s(matrix: &mut Matrix, nid: NodeId, parm: &str, set: i64) {
let p = nid.inp_param(parm).unwrap();
matrix.set_param(p, SAtom::setting(set));
}
#[allow(unused)]
pub fn pset_n(matrix: &mut Matrix, nid: NodeId, parm: &str, v_norm: f32) {
let p = nid.inp_param(parm).unwrap();
matrix.set_param(p, SAtom::param(v_norm));
}
#[allow(unused)]
pub fn pset_d(matrix: &mut Matrix, nid: NodeId, parm: &str, v_denorm: f32) {
let p = nid.inp_param(parm).unwrap();
matrix.set_param(p, SAtom::param(p.norm(v_denorm)));
}
#[allow(unused)]
pub fn pset_n_wait(matrix: &mut Matrix, ne: &mut NodeExecutor, nid: NodeId, parm: &str, v_norm: f32) {
let p = nid.inp_param(parm).unwrap();
matrix.set_param(p, SAtom::param(v_norm));
run_for_ms(ne, 15.0);
}
#[allow(unused)]
pub fn pset_d_wait(matrix: &mut Matrix, ne: &mut NodeExecutor, nid: NodeId, parm: &str, v_denorm: f32) {
let p = nid.inp_param(parm).unwrap();
matrix.set_param(p, SAtom::param(p.norm(v_denorm)));
run_for_ms(ne, 15.0);
}
#[allow(dead_code)]
pub fn save_wav(name: &str, buf: &[f32]) {
let spec = hound::WavSpec {
channels: 1,
sample_rate: SAMPLE_RATE as u32,
bits_per_sample: 16,
sample_format: hound::SampleFormat::Int,
};
let mut writer = hound::WavWriter::create(name, spec).unwrap();
for s in buf.iter() {
let amp = i16::MAX as f32;
writer.write_sample((amp * s) as i16).unwrap();
}
}
pub fn run_no_input(node_exec: &mut hexodsp::nodes::NodeExecutor, seconds: f32) -> (Vec<f32>, Vec<f32>) {
run_realtime_no_input(node_exec, seconds, false)
}
#[allow(dead_code)]
pub fn run_for_ms(node_exec: &mut hexodsp::nodes::NodeExecutor, ms: f32) -> (Vec<f32>, Vec<f32>) {
run_realtime_no_input(node_exec, ms / 1000.0, false)
}
pub fn run_realtime_no_input(node_exec: &mut hexodsp::nodes::NodeExecutor, seconds: f32, sleep_a_bit: bool) -> (Vec<f32>, Vec<f32>) {
node_exec.test_run(seconds, sleep_a_bit)
}
pub fn calc_rms_mimax_each_ms(buf: &[f32], ms: f32) -> Vec<(f32, f32, f32)> {
let ms_samples = ms * SAMPLE_RATE / 1000.0;
let len_ms = ms_samples as usize;
let mut idx = 0;
let mut res = vec![];
loop {
if (idx + len_ms) > buf.len() {
break;
}
let mut max = -1000.0;
let mut min = 1000.0;
for s in buf[idx..(idx + len_ms)].iter() {
max = s.max(max);
min = s.min(min);
}
let rms : f32 =
buf[idx..(idx + len_ms)]
.iter()
.map(|s: &f32| s * s).sum::<f32>()
/ ms_samples;
res.push((rms, min, max));
idx += len_ms;
}
res
}
#[allow(dead_code)]
pub fn run_and_undersample(
node_exec: &mut hexodsp::nodes::NodeExecutor,
run_len_ms: f32, samples: usize) -> Vec<f32>
{
let (out_l, _out_r) = run_no_input(node_exec, run_len_ms / 1000.0);
let sample_interval = out_l.len() / samples;
let mut out_samples = vec![];
for i in 0..samples {
let idx = i * sample_interval;
out_samples.push(out_l[idx]);
}
out_samples
}
#[allow(dead_code)]
pub fn run_and_get_each_rms_mimax(
node_exec: &mut hexodsp::nodes::NodeExecutor,
len_ms: f32) -> Vec<(f32, f32, f32)>
{
let (out_l, _out_r) = run_no_input(node_exec, (len_ms * 3.0) / 1000.0);
calc_rms_mimax_each_ms(&out_l[..], len_ms)
}
#[allow(dead_code)]
pub fn run_and_get_first_rms_mimax(
node_exec: &mut hexodsp::nodes::NodeExecutor,
len_ms: f32) -> (f32, f32, f32)
{
let (out_l, _out_r) = run_no_input(node_exec, (len_ms * 3.0) / 1000.0);
let rms_mimax = calc_rms_mimax_each_ms(&out_l[..], len_ms);
rms_mimax[0]
}
#[allow(unused)]
pub fn run_and_get_l_rms_mimax(
node_exec: &mut hexodsp::nodes::NodeExecutor,
len_ms: f32) -> (f32, f32, f32)
{
let (out_l, _out_r) = run_no_input(node_exec, (len_ms * 3.0) / 1000.0);
let rms_mimax = calc_rms_mimax_each_ms(&out_l[..], len_ms);
rms_mimax[1]
}
#[allow(unused)]
pub fn run_and_get_counted_freq(
node_exec: &mut hexodsp::nodes::NodeExecutor, ms: f32)
-> f64
{
let (out_l, _out_r) =
// +0.1 here for some extra samples
// this is just for tuning the frequency counter, so that it detects
// the last swing correctly. It's probably wrong, but the results
// match up better this way.
run_no_input(node_exec, (ms + 0.1) / 1000.0);
let mut zero_trans = 0;
let mut last_val = 0.0;
for s in out_l.iter() {
if last_val >= 0.0 && *s < 0.0 {
zero_trans += 1;
} else if last_val <= 0.0 && *s > 0.0 {
zero_trans += 1;
}
last_val = *s;
}
println!("SAMPLES: {}", out_l.len());
println!("ZERO TRANS: {}", zero_trans);
let trans_per_sample =
// substract the extra samples applied earlier.
(zero_trans as f64) / ((out_l.len() - 4) as f64);
trans_per_sample * 44100.0 * 0.5
}
#[allow(unused)]
pub fn run_and_get_fft4096(
node_exec: &mut hexodsp::nodes::NodeExecutor,
thres: u32,
offs_ms: f32) -> Vec<(u16, u32)>
{
let min_samples_for_fft = 4096.0;
let offs_samples = (offs_ms * (SAMPLE_RATE / 1000.0)).ceil();
let min_len_samples =
offs_samples
// 2.0 * for safety margin
+ 2.0 * min_samples_for_fft;
let run_len_s = min_len_samples / SAMPLE_RATE;
let (mut out_l, _out_r) = run_no_input(node_exec, run_len_s);
fft_thres_at_ms(&mut out_l[..], FFT::F4096, thres, offs_ms)
}
#[allow(unused)]
pub fn run_and_get_fft4096_2(
node_exec: &mut hexodsp::nodes::NodeExecutor,
thres: u32) -> Vec<(u16, u32)>
{
let min_samples_for_fft = 4096.0;
let min_len_samples = 2.0 * min_samples_for_fft;
let run_len_s = min_len_samples / SAMPLE_RATE;
let (mut out_l, _out_r) = run_no_input(node_exec, run_len_s);
fft(&mut out_l[..], FFT::F4096, thres)
}
#[allow(unused)]
pub fn calc_exp_avg_buckets4096(fft: &[(u16, u32)]) -> Vec<(u16, u32)> {
let mut avg = vec![];
let mut last_n = [0; 256];
let mut p = 0;
let mut cur_len = 2;
for (i, (fq, lvl)) in fft.iter().enumerate() {
last_n[p] = *lvl;
p += 1;
if p >= cur_len {
avg.push((
*fq,
last_n
.iter()
.take(cur_len)
.map(|x| *x)
.sum::<u32>()
/ (cur_len as u32)));
p = 0;
}
if i % 16 == 0 {
cur_len += 2;
if cur_len > last_n.len() {
cur_len = last_n.len();
}
//d// println!("len={}", cur_len);
}
}
avg
}
#[allow(unused)]
pub fn avg_fft_freqs(round_by: f32, ranges: &[u16], fft: &[(u16, u32)]) -> Vec<(u16, u32)> {
let mut from = 0;
let mut out = vec![];
for rng in ranges.iter() {
out.push(
(from,
((avg_fft_range(from, *rng, fft)
/ round_by)
.floor() * round_by)
as u32));
from = *rng;
}
out
}
#[allow(unused)]
pub fn avg_fft_range(from_freq: u16, to_freq: u16, fft: &[(u16, u32)]) -> f32 {
let mut count = 0;
let mut sum = 0;
for (fq, lvl) in fft.iter() {
if from_freq <= *fq && *fq < to_freq {
sum += *lvl;
count += 1;
}
}
sum as f32 / count as f32
}
#[allow(unused)]
pub fn run_and_get_fft512(
node_exec: &mut hexodsp::nodes::NodeExecutor,
thres: u32,
offs_ms: f32) -> Vec<(u16, u32)>
{
let min_samples_for_fft = 512.0;
let offs_samples = (offs_ms * (SAMPLE_RATE / 1000.0)).ceil();
let min_len_samples =
offs_samples
// 2.0 * for safety margin
+ 2.0 * min_samples_for_fft;
let run_len_s = min_len_samples / SAMPLE_RATE;
let (mut out_l, _out_r) = run_no_input(node_exec, run_len_s);
fft_thres_at_ms(&mut out_l[..], FFT::F512, thres, offs_ms)
}
#[allow(unused)]
pub fn run_and_get_fft4096_now(
node_exec: &mut hexodsp::nodes::NodeExecutor,
thres: u32) -> Vec<(u16, u32)>
{
let min_samples_for_fft = 4096.0 * 1.5; // 1.5 for some extra margin
let run_len_s = min_samples_for_fft / SAMPLE_RATE;
let (mut out_l, _out_r) = run_no_input(node_exec, run_len_s);
fft_thres_at_ms(&mut out_l[..], FFT::F4096, thres, 0.0)
}
#[allow(unused)]
pub enum FFT {
F16,
F32,
F64,
F128,
F512,
F1024,
F2048,
F4096,
F8192,
F16384,
F65535,
}
impl FFT {
pub fn size(&self) -> usize {
match self {
FFT::F16 => 16,
FFT::F32 => 32,
FFT::F64 => 64,
FFT::F128 => 128,
FFT::F512 => 512,
FFT::F1024 => 1024,
FFT::F2048 => 2048,
FFT::F4096 => 4096,
FFT::F8192 => 8192,
FFT::F16384 => 16384,
FFT::F65535 => 65535,
}
}
}
pub fn fft_thres_at_ms(buf: &mut [f32], size: FFT, amp_thres: u32, ms_idx: f32) -> Vec<(u16, u32)> {
let ms_sample_offs = ms_idx * (SAMPLE_RATE / 1000.0);
let fft_nbins = size.size();
let len = fft_nbins;
let idx = ms_sample_offs as usize;
if (idx + len) > buf.len() {
return vec![];
}
fft(&mut buf[idx..(idx + len)], size, amp_thres)
}
pub fn fft(buf: &mut [f32], size: FFT, amp_thres: u32) -> Vec<(u16, u32)> {
let len = size.size();
let mut res = vec![];
if len > buf.len() {
return res;
}
// Hann window:
for (i, s) in buf[0..len].iter_mut().enumerate() {
let w =
0.5
* (1.0
- ((2.0 * std::f32::consts::PI * i as f32)
/ (len as f32 - 1.0))
.cos());
*s *= w;
}
use rustfft::{FftPlanner, num_complex::Complex};
let mut complex_buf =
buf.iter()
.map(|s| Complex { re: *s, im: 0.0 })
.collect::<Vec<Complex<f32>>>();
let mut p = FftPlanner::<f32>::new();
let fft = p.plan_fft_forward(len);
fft.process(&mut complex_buf[0..len]);
let amplitudes: Vec<_> =
complex_buf[0..len].iter().map(|c| c.norm() as u32).collect();
// println!("fft: {:?}", &complex_buf[0..len]);
for (i, amp) in amplitudes.iter().enumerate() {
if *amp >= amp_thres {
let freq = (i as f32 * SAMPLE_RATE) / len as f32;
if freq > 22050.0 {
// no freqency images above nyquist...
continue;
}
// println!("{:6.0} {}", freq, *amp);
res.push((freq.round() as u16, *amp));
}
}
res
}