Added test files. Debugged quite some things. SLM needs still tests on statistics. Bugfix in frequency weighting filters. Added a lot of wrappers for Python calls.
This commit is contained in:
parent
90c7f0eb37
commit
7092ef4717
1
.gitignore
vendored
1
.gitignore
vendored
@ -5,3 +5,4 @@ python/lasprs/_lasprs*
|
|||||||
.venv
|
.venv
|
||||||
.vscode/launch.json
|
.vscode/launch.json
|
||||||
.vscode
|
.vscode
|
||||||
|
examples_py/.ipynb_checkpoints
|
||||||
|
135
examples_py/test_filterbank.ipynb
Normal file
135
examples_py/test_filterbank.ipynb
Normal file
@ -0,0 +1,135 @@
|
|||||||
|
{
|
||||||
|
"cells": [
|
||||||
|
{
|
||||||
|
"cell_type": "markdown",
|
||||||
|
"id": "bcd07c1e-6722-44c7-b162-267f1341c2fe",
|
||||||
|
"metadata": {},
|
||||||
|
"source": [
|
||||||
|
"# Standard filterbank frequency response and impulse response"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "f0c4d6ef-69b2-4b5c-92fd-d987a11f3cbd",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"# Prerequisites, uncomment below in case of errors. Also for ipympl, restart Jupyter Lab if it was not installed\n",
|
||||||
|
"\n",
|
||||||
|
"!cd .. && maturin develop -F python-bindings\n",
|
||||||
|
"#!pip install ipympl scipy matplotlib"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "23b45cf6-85bc-4eaa-8f27-77c5b354e772",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"# If this does not work, install ipympl and reboot Jupyter Lab\n",
|
||||||
|
"%matplotlib widget"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "96090ac9-2033-411d-8e45-0c6b375a268b",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"import numpy as np\n",
|
||||||
|
"import matplotlib.pyplot as plt\n",
|
||||||
|
"from numpy import log10, sqrt, exp, pi\n",
|
||||||
|
"from lasprs._lasprs import StandardFilterDescriptor, SLMSettings, FreqWeighting, TimeWeighting, SLM\n",
|
||||||
|
"def level(a):\n",
|
||||||
|
" return 20*np.log10(np.abs(a))"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "52343e3d-63ef-4fc2-ad7e-85ca124efdac",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"freq = np.logspace(log10(2), log10(2e4), 200)\n",
|
||||||
|
"octaves = StandardFilterDescriptor.genFilterSetInRange(1, 10, 16e3, False)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "2788f062-34ab-4573-9849-459fdccc97d3",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"hs = [ o.genFilter().tf(0,freq) for o in octaves]\n",
|
||||||
|
"names = [str(o) for o in octaves]"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "3366f635-29fe-4757-ab0c-9d4929f8cddb",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"plt.figure()\n",
|
||||||
|
"# plt.subplot(211)\n",
|
||||||
|
"plt.title('Frequency response of filters - magnitude')\n",
|
||||||
|
"\n",
|
||||||
|
"for h in hs:\n",
|
||||||
|
" plt.semilogx(freq, level(h))\n",
|
||||||
|
"plt.legend(names)\n",
|
||||||
|
"plt.ylabel('Magnitude [dB]')\n",
|
||||||
|
"plt.ylim(-50, 1)\n",
|
||||||
|
"plt.xlabel('Freq. [Hz]')"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "b242187a-3449-4c30-8e26-28e75d59c414",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"fs = 48000\n",
|
||||||
|
"tend = 0.01\n",
|
||||||
|
"t = np.linspace(0, tend, int(tend*fs), endpoint=False)\n",
|
||||||
|
"impulse = np.zeros(t.size)\n",
|
||||||
|
"impulse[0] = 1\n",
|
||||||
|
"plt.figure()\n",
|
||||||
|
"plt.title('Filter impulse response')\n",
|
||||||
|
"for o in octaves:\n",
|
||||||
|
" i = o.genFilter().bilinear(fs).filter(impulse)\n",
|
||||||
|
" plt.plot(t, i)\n",
|
||||||
|
"plt.legend(names, loc='upper right')\n",
|
||||||
|
"plt.ylabel('Filter output [-]')\n",
|
||||||
|
"plt.xlabel('Time [s]')"
|
||||||
|
]
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"metadata": {
|
||||||
|
"kernelspec": {
|
||||||
|
"display_name": "Python 3 (ipykernel)",
|
||||||
|
"language": "python",
|
||||||
|
"name": "python3"
|
||||||
|
},
|
||||||
|
"language_info": {
|
||||||
|
"codemirror_mode": {
|
||||||
|
"name": "ipython",
|
||||||
|
"version": 3
|
||||||
|
},
|
||||||
|
"file_extension": ".py",
|
||||||
|
"mimetype": "text/x-python",
|
||||||
|
"name": "python",
|
||||||
|
"nbconvert_exporter": "python",
|
||||||
|
"pygments_lexer": "ipython3",
|
||||||
|
"version": "3.10.12"
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"nbformat": 4,
|
||||||
|
"nbformat_minor": 5
|
||||||
|
}
|
110
examples_py/test_freqweighting.ipynb
Normal file
110
examples_py/test_freqweighting.ipynb
Normal file
@ -0,0 +1,110 @@
|
|||||||
|
{
|
||||||
|
"cells": [
|
||||||
|
{
|
||||||
|
"cell_type": "markdown",
|
||||||
|
"id": "bcd07c1e-6722-44c7-b162-267f1341c2fe",
|
||||||
|
"metadata": {},
|
||||||
|
"source": [
|
||||||
|
"# Frequency weighting plots"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "f0c4d6ef-69b2-4b5c-92fd-d987a11f3cbd",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"# Prerequisites, uncomment below in case of errors. Also for ipympl, restart Jupyter Lab if it was not installed\n",
|
||||||
|
"\n",
|
||||||
|
"!cd .. && maturin develop -F python-bindings\n",
|
||||||
|
"#!pip install ipympl scipy matplotlib"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "23b45cf6-85bc-4eaa-8f27-77c5b354e772",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"# If this does not work, install ipympl and reboot Jupyter Lab\n",
|
||||||
|
"%matplotlib widget"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "96090ac9-2033-411d-8e45-0c6b375a268b",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"import numpy as np\n",
|
||||||
|
"import matplotlib.pyplot as plt\n",
|
||||||
|
"from numpy import log10, sqrt, exp, pi\n",
|
||||||
|
"from lasprs._lasprs import ZPKModel, FreqWeighting\n",
|
||||||
|
"def level(a):\n",
|
||||||
|
" return 20*np.log10(np.abs(a))"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "52343e3d-63ef-4fc2-ad7e-85ca124efdac",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"A = ZPKModel.freqWeightingFilter(FreqWeighting.A)\n",
|
||||||
|
"C = ZPKModel.freqWeightingFilter(FreqWeighting.C)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "7d3cc799-4809-4ea4-8570-89ffbb751a5a",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"freq = np.logspace(log10(2), log10(2e4), 200)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "ab22d9cc-03ae-4d58-b730-599b659f9dc0",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"plt.figure()\n",
|
||||||
|
"plt.title('Frequency response of weighting filters - magnitude')\n",
|
||||||
|
"plt.semilogx(freq, level(A.tf(0, freq)))\n",
|
||||||
|
"plt.semilogx(freq, level(C.tf(0, freq)))\n",
|
||||||
|
"plt.legend(['A-weighting', 'Z-weighting'])\n",
|
||||||
|
"plt.ylabel('Magnitude [dB]')\n",
|
||||||
|
"plt.ylim(-50, 3)\n",
|
||||||
|
"plt.xlabel('Freq. [Hz]')"
|
||||||
|
]
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"metadata": {
|
||||||
|
"kernelspec": {
|
||||||
|
"display_name": "Python 3 (ipykernel)",
|
||||||
|
"language": "python",
|
||||||
|
"name": "python3"
|
||||||
|
},
|
||||||
|
"language_info": {
|
||||||
|
"codemirror_mode": {
|
||||||
|
"name": "ipython",
|
||||||
|
"version": 3
|
||||||
|
},
|
||||||
|
"file_extension": ".py",
|
||||||
|
"mimetype": "text/x-python",
|
||||||
|
"name": "python",
|
||||||
|
"nbconvert_exporter": "python",
|
||||||
|
"pygments_lexer": "ipython3",
|
||||||
|
"version": "3.10.12"
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"nbformat": 4,
|
||||||
|
"nbformat_minor": 5
|
||||||
|
}
|
175
examples_py/test_slm1.ipynb
Normal file
175
examples_py/test_slm1.ipynb
Normal file
@ -0,0 +1,175 @@
|
|||||||
|
{
|
||||||
|
"cells": [
|
||||||
|
{
|
||||||
|
"cell_type": "markdown",
|
||||||
|
"id": "bcd07c1e-6722-44c7-b162-267f1341c2fe",
|
||||||
|
"metadata": {},
|
||||||
|
"source": [
|
||||||
|
"# Test Sound Level Meter implementation - 1"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "f0c4d6ef-69b2-4b5c-92fd-d987a11f3cbd",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"# Prerequisites, uncomment below in case of errors. Also for ipympl, restart Jupyter Lab if it was not installed\n",
|
||||||
|
"\n",
|
||||||
|
"!cd .. && maturin develop -F python-bindings\n",
|
||||||
|
"#!pip install ipympl scipy matplotlib"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "23b45cf6-85bc-4eaa-8f27-77c5b354e772",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"# If this does not work, install ipympl and reboot Jupyter Lab\n",
|
||||||
|
"%matplotlib widget"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "96090ac9-2033-411d-8e45-0c6b375a268b",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"import numpy as np\n",
|
||||||
|
"import matplotlib.pyplot as plt\n",
|
||||||
|
"from numpy import log10, sqrt, exp, pi\n",
|
||||||
|
"from lasprs._lasprs import StandardFilterDescriptor, SLMSettings, FreqWeighting, TimeWeighting, SLM\n",
|
||||||
|
"def level(a):\n",
|
||||||
|
" return 20*np.log10(np.abs(a))"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "52343e3d-63ef-4fc2-ad7e-85ca124efdac",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"freq = np.logspace(log10(2), log10(2e4), 200)\n",
|
||||||
|
"octaves = [StandardFilterDescriptor.Overall()]+StandardFilterDescriptor.genOctaveFilterSet(16, 16e3)\n",
|
||||||
|
"\n",
|
||||||
|
"names = [str(o) for o in octaves]"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "ab22d9cc-03ae-4d58-b730-599b659f9dc0",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"fs = 48000\n",
|
||||||
|
"\n",
|
||||||
|
"tw = TimeWeighting.Impulse()\n",
|
||||||
|
"# tw = TimeWeighting.Fast()\n",
|
||||||
|
"N = fs\n",
|
||||||
|
"if tw == TimeWeighting.Slow():\n",
|
||||||
|
" tau = 1.0\n",
|
||||||
|
"elif tw == TimeWeighting.Fast():\n",
|
||||||
|
" tau = 1/8\n",
|
||||||
|
"elif tw == TimeWeighting.Impulse():\n",
|
||||||
|
" tau = 35e-3\n",
|
||||||
|
"\n",
|
||||||
|
"else:\n",
|
||||||
|
" raise NotImplementedError()\n",
|
||||||
|
"settings = SLMSettings(fs, FreqWeighting.Z, tw, filterDescriptors=octaves, Lref=1)\n",
|
||||||
|
"slm = SLM(settings)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "e1960b80-99f9-4744-9b0f-6651ed2693cd",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"inp = np.ones(N)\n",
|
||||||
|
"# inp = np.zeros(N)\n",
|
||||||
|
"inp[0] = 1"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "532539b8-7a5e-47c2-afb1-a72d62c445d4",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"t = np.linspace(0, N/fs, N, endpoint=False)\n",
|
||||||
|
"out = slm.run(inp)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "37af9b78-73b9-4228-a758-d9a5362f79f4",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"len(out[0])"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "4609a63f-e350-4b6b-b898-93c9e6fef804",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"# help(plt.legend)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "f446ac32-4ac2-4e85-87c8-43a06be9fb14",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": [
|
||||||
|
"plt.figure()\n",
|
||||||
|
"for o in out:\n",
|
||||||
|
" plt.plot(t,o)\n",
|
||||||
|
"plt.ylim(-60, 0)\n",
|
||||||
|
"plt.legend(names)"
|
||||||
|
]
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"cell_type": "code",
|
||||||
|
"execution_count": null,
|
||||||
|
"id": "4303e20e-cc61-4d23-a5f0-cab683d55912",
|
||||||
|
"metadata": {},
|
||||||
|
"outputs": [],
|
||||||
|
"source": []
|
||||||
|
}
|
||||||
|
],
|
||||||
|
"metadata": {
|
||||||
|
"kernelspec": {
|
||||||
|
"display_name": "Python 3 (ipykernel)",
|
||||||
|
"language": "python",
|
||||||
|
"name": "python3"
|
||||||
|
},
|
||||||
|
"language_info": {
|
||||||
|
"codemirror_mode": {
|
||||||
|
"name": "ipython",
|
||||||
|
"version": 3
|
||||||
|
},
|
||||||
|
"file_extension": ".py",
|
||||||
|
"mimetype": "text/x-python",
|
||||||
|
"name": "python",
|
||||||
|
"nbconvert_exporter": "python",
|
||||||
|
"pygments_lexer": "ipython3",
|
||||||
|
"version": "3.10.12"
|
||||||
|
}
|
||||||
|
},
|
||||||
|
"nbformat": 4,
|
||||||
|
"nbformat_minor": 5
|
||||||
|
}
|
@ -101,7 +101,7 @@ impl Biquad {
|
|||||||
&mut self,
|
&mut self,
|
||||||
py: Python<'py>,
|
py: Python<'py>,
|
||||||
input: PyArrayLike1<Flt>,
|
input: PyArrayLike1<Flt>,
|
||||||
) -> PyResult<PyArr1Flt<'py>> {
|
) -> Result<PyArr1Flt<'py>> {
|
||||||
Ok(PyArray1::from_vec_bound(py, self.filter(input.as_slice()?)))
|
Ok(PyArray1::from_vec_bound(py, self.filter(input.as_slice()?)))
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
@ -128,12 +128,25 @@ impl Biquad {
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Re-initialize state. *This is an advanced function. You should know what
|
/// Set the state as if the filter converged to the DC value `val`. Note
|
||||||
/// you are doing!*. If not, please use any other function like
|
/// that this only works for filters that do have a D.C. nonzero gain. If
|
||||||
/// [Biquad::reset].
|
/// not, this method will do a kind of divide-by-zero.
|
||||||
pub fn setNextOutputX0(&mut self, out: Flt) {
|
pub fn setToDCValue(&mut self, val: Flt) {
|
||||||
let (b0, b1, b2, a1, a2) = (self.b0, self.b1, self.b2, self.a1, self.a2);
|
// D.C. output is:
|
||||||
let w = out / (b1 + b2 - b0 * (a1 + a2));
|
// y[n] = b0*w[n] + b1*w[n-1] + b2*w[n-2]
|
||||||
|
// and:
|
||||||
|
// w[n] = x[n] - a1*w[n-1] -a2*w[n-2]
|
||||||
|
|
||||||
|
// Assume w[n] = w[n-1] = w[n-2]:
|
||||||
|
// We have:
|
||||||
|
// w[n]*(1+a1+a2) = x
|
||||||
|
// And:
|
||||||
|
// y[n] = (b0+b1+b2) * wn
|
||||||
|
// Hence:
|
||||||
|
// y[n] = sum_b * w
|
||||||
|
// So we should set w to val/sum_b
|
||||||
|
let sumb = self.b0 + self.b1 + self.b2;
|
||||||
|
let w = val / sumb;
|
||||||
self.w1 = w;
|
self.w1 = w;
|
||||||
self.w2 = w;
|
self.w2 = w;
|
||||||
}
|
}
|
||||||
@ -442,6 +455,16 @@ mod test {
|
|||||||
let filtered = ser.filter(&inp);
|
let filtered = ser.filter(&inp);
|
||||||
assert_eq!(&filtered, &inp);
|
assert_eq!(&filtered, &inp);
|
||||||
}
|
}
|
||||||
|
#[test]
|
||||||
|
fn test_setDC() {
|
||||||
|
let mut b = Biquad::firstOrderMovingAverage(7., 1.).unwrap();
|
||||||
|
let dc = 7.8;
|
||||||
|
b.setToDCValue(dc);
|
||||||
|
|
||||||
|
let mut x = dc;
|
||||||
|
b.filter_inout_single(&mut x);
|
||||||
|
assert_abs_diff_eq!(x, dc, epsilon = Flt::EPSILON * 10.)
|
||||||
|
}
|
||||||
|
|
||||||
#[test]
|
#[test]
|
||||||
fn test_firstOrderLowpass() {
|
fn test_firstOrderLowpass() {
|
||||||
@ -502,21 +525,4 @@ mod test {
|
|||||||
// let freq = &[0., 10.,100.,1000., 2000.];
|
// let freq = &[0., 10.,100.,1000., 2000.];
|
||||||
// println!("{:?}", b3.tf(fs, freq));
|
// println!("{:?}", b3.tf(fs, freq));
|
||||||
}
|
}
|
||||||
|
|
||||||
#[test]
|
|
||||||
fn test_setOutput1() {
|
|
||||||
let mut f = Biquad::firstOrderHighPass(10., 1.).unwrap();
|
|
||||||
f.setNextOutputX0(1.0);
|
|
||||||
let mut sample = 0.;
|
|
||||||
f.filter_inout_single(&mut sample);
|
|
||||||
assert_abs_diff_eq!(sample, 1.0);
|
|
||||||
}
|
|
||||||
#[test]
|
|
||||||
fn test_setOutput2() {
|
|
||||||
let mut f = Biquad::bilinear_zpk(1.0, None, Some(PoleOrZero::Real1(-1.)), None, None);
|
|
||||||
f.setNextOutputX0(4.2);
|
|
||||||
let mut sample = 0.;
|
|
||||||
f.filter_inout_single(&mut sample);
|
|
||||||
assert_abs_diff_eq!(sample, 4.2, epsilon = 1e-6);
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
|
@ -1,4 +1,4 @@
|
|||||||
use crate::{Flt, ZPKModel};
|
use crate::{config::*, ZPKModel};
|
||||||
use anyhow::{anyhow, bail, Result};
|
use anyhow::{anyhow, bail, Result};
|
||||||
use num::{traits::float, Float};
|
use num::{traits::float, Float};
|
||||||
use rayon::iter::Filter;
|
use rayon::iter::Filter;
|
||||||
@ -100,6 +100,7 @@ pub const FREQ_REF: Flt = 1000.;
|
|||||||
/// # Ok(())
|
/// # Ok(())
|
||||||
/// # }
|
/// # }
|
||||||
/// ```
|
/// ```
|
||||||
|
#[cfg_attr(feature = "python-bindings", pyclass)]
|
||||||
#[derive(PartialEq, Clone, Debug)]
|
#[derive(PartialEq, Clone, Debug)]
|
||||||
pub struct StandardFilterDescriptor {
|
pub struct StandardFilterDescriptor {
|
||||||
/// b and x. Bandwidth and offset w.r.t. reference frequency.
|
/// b and x. Bandwidth and offset w.r.t. reference frequency.
|
||||||
@ -128,12 +129,12 @@ impl StandardFilterDescriptor {
|
|||||||
// are valid.
|
// are valid.
|
||||||
fn check_fmid_in_range(&self) -> Result<()> {
|
fn check_fmid_in_range(&self) -> Result<()> {
|
||||||
if let Some(fm) = self.fm() {
|
if let Some(fm) = self.fm() {
|
||||||
if fm < MIN_MIDBAND_FREQ {
|
if fm < MIN_MIDBAND_FREQ / 2. {
|
||||||
bail!(
|
bail!(
|
||||||
"Invalid x. Computed filter center frequency is {} Hz, which is too low. Lowest allowed is {} Hz",
|
"Invalid x. Computed filter center frequency is {} Hz, which is too low. Lowest allowed is {} Hz",
|
||||||
fm, MIN_MIDBAND_FREQ
|
fm, MIN_MIDBAND_FREQ
|
||||||
)
|
)
|
||||||
} else if fm > 20e3 {
|
} else if fm > 25e3 {
|
||||||
bail!(
|
bail!(
|
||||||
"Invalid x. Computed filter center frequency is {} Hz, which is too high. Highest allowed is {} Hz",
|
"Invalid x. Computed filter center frequency is {} Hz, which is too high. Highest allowed is {} Hz",
|
||||||
fm, MAX_MIDBAND_FREQ
|
fm, MAX_MIDBAND_FREQ
|
||||||
@ -216,9 +217,9 @@ impl StandardFilterDescriptor {
|
|||||||
};
|
};
|
||||||
|
|
||||||
loop {
|
loop {
|
||||||
|
// Get midband. Assuming we are not overall if we arrive
|
||||||
|
// here in the loop
|
||||||
let fm = desc.fm().unwrap();
|
let fm = desc.fm().unwrap();
|
||||||
// eprintln!("Fmid: {fm:.2e}");
|
|
||||||
// eprintln!("desc: {desc:#?}");
|
|
||||||
let ord = f_in_range(&desc);
|
let ord = f_in_range(&desc);
|
||||||
// Bands for midband frequencies are a bit wider here
|
// Bands for midband frequencies are a bit wider here
|
||||||
if fm < MIN_MIDBAND_FREQ - 3. || fm > MAX_MIDBAND_FREQ * 1.1 {
|
if fm < MIN_MIDBAND_FREQ - 3. || fm > MAX_MIDBAND_FREQ * 1.1 {
|
||||||
@ -284,24 +285,24 @@ impl StandardFilterDescriptor {
|
|||||||
/// overall, `b=1` is one octave, `b=3`` is one-third, etc.
|
/// overall, `b=1` is one octave, `b=3`` is one-third, etc.
|
||||||
/// - `xmin` - Band designator of lowest band. Midband frequency can be computed as [FREQ_REF]*[G]^(`xmin/b`)
|
/// - `xmin` - Band designator of lowest band. Midband frequency can be computed as [FREQ_REF]*[G]^(`xmin/b`)
|
||||||
/// - `xmax` - Band designator of lowest band. Midband frequency can be computed as [FREQ_REF]*[G]^(`xmax/b`)
|
/// - `xmax` - Band designator of lowest band. Midband frequency can be computed as [FREQ_REF]*[G]^(`xmax/b`)
|
||||||
/// - `include_overall` - If `true`, adds an overall filter (a no-op) as the last designator in the list
|
/// - `append_overall` - If `true`, adds an overall filter (a no-op) as the last designator in the list
|
||||||
pub fn genFilterSetByDesignator(
|
pub fn genFilterSetByDesignator(
|
||||||
b: u32,
|
b: u32,
|
||||||
xmin: i32,
|
xmin: i32,
|
||||||
xmax: i32,
|
xmax: i32,
|
||||||
include_overall: bool,
|
append_overall: bool,
|
||||||
) -> Result<Vec<Self>> {
|
) -> Result<Vec<Self>> {
|
||||||
if xmin > xmax {
|
if xmin > xmax {
|
||||||
bail!("xmin should be <= xmax");
|
bail!("xmin should be <= xmax");
|
||||||
}
|
}
|
||||||
let cap = (xmax - xmin) as usize + if include_overall { 1 } else { 0 };
|
let cap = (xmax - xmin) as usize + if append_overall { 1 } else { 0 };
|
||||||
let mut res = Vec::with_capacity(cap);
|
let mut res = Vec::with_capacity(cap);
|
||||||
|
|
||||||
for x in xmin..=xmax {
|
for x in xmin..=xmax {
|
||||||
res.push(StandardFilterDescriptor::build(b, x)?);
|
res.push(StandardFilterDescriptor::build(b, x)?);
|
||||||
}
|
}
|
||||||
|
|
||||||
if include_overall {
|
if append_overall {
|
||||||
res.push(StandardFilterDescriptor::Overall()?)
|
res.push(StandardFilterDescriptor::Overall()?)
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -314,16 +315,17 @@ impl StandardFilterDescriptor {
|
|||||||
///
|
///
|
||||||
/// # Other args
|
/// # Other args
|
||||||
///
|
///
|
||||||
/// - `include_overall` - If `true`, adds an overall filter (a no-op) as the
|
/// - `append_overall` - If `true`, adds an overall filter (a no-op) as the
|
||||||
pub fn genFilterSetInRange(
|
/// last filter in the list.
|
||||||
|
pub fn genFilterSetForRange(
|
||||||
b: u32,
|
b: u32,
|
||||||
fl: Flt,
|
fl: Flt,
|
||||||
fu: Flt,
|
fu: Flt,
|
||||||
include_overall: bool,
|
append_overall: bool,
|
||||||
) -> Result<Vec<Self>> {
|
) -> Result<Vec<Self>> {
|
||||||
let xmin = StandardFilterDescriptor::filterForFreq(b, fl)?.x;
|
let xmin = StandardFilterDescriptor::filterForFreq(b, fl)?.x;
|
||||||
let xmax = StandardFilterDescriptor::filterForFreq(b, fu)?.x;
|
let xmax = StandardFilterDescriptor::filterForFreq(b, fu)?.x;
|
||||||
StandardFilterDescriptor::genFilterSetByDesignator(b, xmin, xmax, include_overall)
|
StandardFilterDescriptor::genFilterSetByDesignator(b, xmin, xmax, append_overall)
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Returns the midband frequency in \[Hz\]
|
/// Returns the midband frequency in \[Hz\]
|
||||||
@ -338,7 +340,7 @@ impl StandardFilterDescriptor {
|
|||||||
}
|
}
|
||||||
|
|
||||||
/// Cuton frequency and cut-off frequency, in \[Hz\].
|
/// Cuton frequency and cut-off frequency, in \[Hz\].
|
||||||
/// Returns none if it does not apply, for [FilterDescriptor::Overall].
|
/// Returns none if it does not apply, for [StandardFilterDescriptor::Overall].
|
||||||
pub fn fl_fh(&self) -> Option<(Flt, Flt)> {
|
pub fn fl_fh(&self) -> Option<(Flt, Flt)> {
|
||||||
match self.b {
|
match self.b {
|
||||||
0 => None,
|
0 => None,
|
||||||
@ -435,6 +437,22 @@ impl TryFrom<&str> for ThirdOctaveBandDescriptor {
|
|||||||
})
|
})
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
impl TryFrom<Flt> for ThirdOctaveBandDescriptor {
|
||||||
|
type Error = anyhow::Error;
|
||||||
|
fn try_from(value: Flt) -> Result<Self, Self::Error> {
|
||||||
|
Ok(ThirdOctaveBandDescriptor {
|
||||||
|
x: StandardFilterDescriptor::filterForFreq(3, value)?.x,
|
||||||
|
})
|
||||||
|
}
|
||||||
|
}
|
||||||
|
impl TryFrom<Flt> for OctaveBandDescriptor {
|
||||||
|
type Error = anyhow::Error;
|
||||||
|
fn try_from(value: Flt) -> Result<Self, Self::Error> {
|
||||||
|
Ok(OctaveBandDescriptor {
|
||||||
|
x: StandardFilterDescriptor::filterForFreq(1, value)?.x,
|
||||||
|
})
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
trait BandDescriptor {
|
trait BandDescriptor {
|
||||||
fn name(&self) -> Cow<'static, str>;
|
fn name(&self) -> Cow<'static, str>;
|
||||||
@ -460,11 +478,58 @@ impl BandDescriptor for ThirdOctaveBandDescriptor {
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// impl Into<OctaveBandDescriptor> for &str {
|
#[cfg_attr(feature = "python-bindings", pymethods)]
|
||||||
// fn into(self) -> OctaveBandDescriptor {
|
impl StandardFilterDescriptor {
|
||||||
// OctaveBandDescriptor{x: self}
|
#[pyo3(name = "genFilter")]
|
||||||
// }
|
fn genFilter_py(&self) -> ZPKModel {
|
||||||
// }
|
self.genFilter()
|
||||||
|
}
|
||||||
|
#[staticmethod]
|
||||||
|
#[pyo3(name = "Overall")]
|
||||||
|
fn Overall_py() -> StandardFilterDescriptor {
|
||||||
|
StandardFilterDescriptor::Overall().unwrap()
|
||||||
|
}
|
||||||
|
|
||||||
|
#[staticmethod]
|
||||||
|
#[pyo3(name = "genThirdOctaveFilterSet")]
|
||||||
|
fn genThirdOctaveFilterSet_py(fmin: Flt, fmax: Flt) -> PyResult<Vec<StandardFilterDescriptor>> {
|
||||||
|
Ok(Self::genThirdOctaveFilterSet(fmin, fmax)?)
|
||||||
|
}
|
||||||
|
|
||||||
|
#[staticmethod]
|
||||||
|
#[pyo3(name = "genOctaveFilterSet")]
|
||||||
|
fn genOctaveFilterSetFromFreq(fmin: Flt, fmax: Flt) -> PyResult<Vec<StandardFilterDescriptor>> {
|
||||||
|
Ok(Self::genOctaveFilterSet(fmin, fmax)?)
|
||||||
|
}
|
||||||
|
|
||||||
|
#[staticmethod]
|
||||||
|
fn genThirdOctaveFilterSetFromFreq(
|
||||||
|
fmin: Flt,
|
||||||
|
fmax: Flt,
|
||||||
|
) -> PyResult<Vec<StandardFilterDescriptor>> {
|
||||||
|
Ok(Self::genThirdOctaveFilterSet(fmin, fmax)?)
|
||||||
|
}
|
||||||
|
|
||||||
|
fn __repr__(&self) -> String {
|
||||||
|
format!{"{:#?}", self}
|
||||||
|
}
|
||||||
|
|
||||||
|
fn __str__(&self) -> String {
|
||||||
|
self.name().into()
|
||||||
|
}
|
||||||
|
|
||||||
|
#[staticmethod]
|
||||||
|
#[pyo3(name = "genFilterSetInRange")]
|
||||||
|
fn genFilterSetInRange_py(
|
||||||
|
b: u32,
|
||||||
|
fmin: Flt,
|
||||||
|
fmax: Flt,
|
||||||
|
append_overall: bool,
|
||||||
|
) -> PyResult<Vec<StandardFilterDescriptor>> {
|
||||||
|
Ok(Self::genFilterSetForRange(b, fmin, fmax, append_overall)?)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
#[cfg(test)]
|
#[cfg(test)]
|
||||||
mod test {
|
mod test {
|
||||||
use super::*;
|
use super::*;
|
||||||
|
@ -89,9 +89,13 @@ pub struct ZPKModel {
|
|||||||
}
|
}
|
||||||
impl Default for ZPKModel {
|
impl Default for ZPKModel {
|
||||||
fn default() -> Self {
|
fn default() -> Self {
|
||||||
ZPKModel{z:vec![], p:vec![], k: 1.0, fwarp: None}
|
ZPKModel {
|
||||||
|
z: vec![],
|
||||||
|
p: vec![],
|
||||||
|
k: 1.0,
|
||||||
|
fwarp: None,
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
}
|
}
|
||||||
impl<'a, T: AsArray<'a, Flt>> TransferFunction<'a, T> for ZPKModel {
|
impl<'a, T: AsArray<'a, Flt>> TransferFunction<'a, T> for ZPKModel {
|
||||||
fn tf(&self, _fs: Flt, freq: T) -> Ccol {
|
fn tf(&self, _fs: Flt, freq: T) -> Ccol {
|
||||||
@ -155,6 +159,12 @@ impl ZPKModel {
|
|||||||
ZPKModel::butter(spec)
|
ZPKModel::butter(spec)
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#[pyo3(name = "freqWeightingFilter")]
|
||||||
|
#[staticmethod]
|
||||||
|
fn freqWeightingFilter_py(fw: FreqWeighting) ->ZPKModel {
|
||||||
|
Self::freqWeightingFilter(fw)
|
||||||
|
}
|
||||||
|
|
||||||
fn __repr__(&self) -> String {
|
fn __repr__(&self) -> String {
|
||||||
format!("{self:?}")
|
format!("{self:?}")
|
||||||
}
|
}
|
||||||
@ -171,6 +181,12 @@ impl ZPKModel {
|
|||||||
let res = PyArray1::from_array_bound(py, &self.tf(fs, freq));
|
let res = PyArray1::from_array_bound(py, &self.tf(fs, freq));
|
||||||
Ok(res)
|
Ok(res)
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/// See: [ZPKModel::tf]
|
||||||
|
#[pyo3(name = "bilinear")]
|
||||||
|
fn bilinear_py(&self, fs: Flt) -> SeriesBiquad {
|
||||||
|
self.bilinear(fs)
|
||||||
|
}
|
||||||
}
|
}
|
||||||
impl ZPKModel {
|
impl ZPKModel {
|
||||||
/// Creata a new ZPK model, with give list of poles and zeros, and gain
|
/// Creata a new ZPK model, with give list of poles and zeros, and gain
|
||||||
@ -538,20 +554,22 @@ impl ZPKModel {
|
|||||||
let fHsq: Flt = fH.powi(2);
|
let fHsq: Flt = fH.powi(2);
|
||||||
let frsq: Flt = fr.powi(2);
|
let frsq: Flt = fr.powi(2);
|
||||||
let fA = num::Float::powf(10., 2.45);
|
let fA = num::Float::powf(10., 2.45);
|
||||||
let D = num::Float::powf(2., 0.5);
|
let D = Flt::sqrt(0.5);
|
||||||
|
|
||||||
let b = (1. / (1. - D)) * (frsq + fLsq * fHsq / frsq - D * (fLsq + fHsq));
|
let b = (1. / (1. - D)) * (frsq + fLsq * fHsq / frsq - D * (fLsq + fHsq));
|
||||||
let c = fLsq * fHsq;
|
let c = fLsq * fHsq;
|
||||||
let f2 = (3. - sq5) / 2. * fA;
|
let f2 = (3. - sq5) / 2. * fA;
|
||||||
let f3 = (3. + sq5) / 2. * fA;
|
let f3 = (3. + sq5) / 2. * fA;
|
||||||
|
|
||||||
let f1 = ((-b - (b.powi(2) - 4. * c).sqrt()) / 2.).sqrt();
|
let sqrtfac = (b.powi(2) - 4. * c).sqrt();
|
||||||
let f4 = ((-b + (b.powi(2) - 4. * c).sqrt()) / 2.).sqrt();
|
let f1 = ((-b - sqrtfac) / 2.).sqrt();
|
||||||
|
let f4 = ((-b + sqrtfac) / 2.).sqrt();
|
||||||
|
|
||||||
let p1 = 2. * pi * f1;
|
let p1 = 2. * pi * f1;
|
||||||
let p2 = 2. * pi * f2;
|
let p2 = 2. * pi * f2;
|
||||||
let p3 = 2. * pi * f3;
|
let p3 = 2. * pi * f3;
|
||||||
let p4 = 2. * pi * f4;
|
let p4 = 2. * pi * f4;
|
||||||
|
println!("{b} {p1}, {p2}, {p3}, {p4}");
|
||||||
|
|
||||||
let (zeros, poles) = match wt {
|
let (zeros, poles) = match wt {
|
||||||
FreqWeighting::Z => {
|
FreqWeighting::Z => {
|
||||||
@ -589,3 +607,14 @@ pub enum PoleOrZero {
|
|||||||
/// Single zero / pole
|
/// Single zero / pole
|
||||||
Real1(Flt),
|
Real1(Flt),
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#[cfg(test)]
|
||||||
|
mod test{
|
||||||
|
use super::ZPKModel;
|
||||||
|
|
||||||
|
|
||||||
|
#[test]
|
||||||
|
fn test_A() {
|
||||||
|
let Aw = ZPKModel::freqWeightingFilter(crate::FreqWeighting::A);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
@ -57,6 +57,11 @@ fn lasprs(m: &Bound<'_, PyModule>) -> PyResult<()> {
|
|||||||
m.add_class::<siggen::Siggen>()?;
|
m.add_class::<siggen::Siggen>()?;
|
||||||
m.add_class::<filter::FilterSpec>()?;
|
m.add_class::<filter::FilterSpec>()?;
|
||||||
m.add_class::<filter::ZPKModel>()?;
|
m.add_class::<filter::ZPKModel>()?;
|
||||||
|
m.add_class::<filter::StandardFilterDescriptor>()?;
|
||||||
|
m.add_class::<slm::TimeWeighting>()?;
|
||||||
|
m.add_class::<ps::FreqWeighting>()?;
|
||||||
|
m.add_class::<slm::SLMSettings>()?;
|
||||||
|
m.add_class::<slm::SLM>()?;
|
||||||
|
|
||||||
Ok(())
|
Ok(())
|
||||||
}
|
}
|
||||||
|
@ -1,6 +1,8 @@
|
|||||||
|
use crate::config::*;
|
||||||
use strum_macros::{Display, EnumMessage};
|
use strum_macros::{Display, EnumMessage};
|
||||||
/// Sound level frequency weighting type (A, C, Z)
|
/// Sound level frequency weighting type (A, C, Z)
|
||||||
#[derive(Display, Debug, EnumMessage, Default, Clone)]
|
#[cfg_attr(feature = "python-bindings", pyclass(eq, eq_int))]
|
||||||
|
#[derive(Display, Debug, EnumMessage, Default, Clone, PartialEq)]
|
||||||
pub enum FreqWeighting {
|
pub enum FreqWeighting {
|
||||||
/// A-weighting
|
/// A-weighting
|
||||||
A,
|
A,
|
||||||
|
@ -1,18 +1,22 @@
|
|||||||
use super::{TimeWeighting, SLM_MAX_CHANNELS};
|
use super::{TimeWeighting, SLM, SLM_MAX_CHANNELS};
|
||||||
use crate::{filter::StandardFilterDescriptor, Flt, FreqWeighting};
|
use crate::{config::*, filter::StandardFilterDescriptor, Flt, FreqWeighting};
|
||||||
use anyhow::Result;
|
use anyhow::Result;
|
||||||
use clap::builder;
|
use clap::builder;
|
||||||
use derive_builder::Builder;
|
use derive_builder::Builder;
|
||||||
use smallvec::{smallvec, SmallVec};
|
use smallvec::{smallvec, SmallVec};
|
||||||
|
|
||||||
|
const Lref_default: Flt = 20e-6;
|
||||||
|
|
||||||
/// Settings used to create a Sound Level Meter.
|
/// Settings used to create a Sound Level Meter.
|
||||||
#[derive(Builder, Clone)]
|
#[derive(Builder, Clone)]
|
||||||
#[builder(setter(into))]
|
#[builder(setter(into))]
|
||||||
|
#[cfg_attr(feature = "python-bindings", pyclass)]
|
||||||
pub struct SLMSettings {
|
pub struct SLMSettings {
|
||||||
/// Sampling frequency in \[Hz\]
|
/// Sampling frequency in \[Hz\]
|
||||||
pub fs: Flt,
|
pub fs: Flt,
|
||||||
/// Reference level, in units of measured quantity. For sound pressure in
|
/// Reference level, in units of measured quantity. For sound pressure in
|
||||||
/// air, this is typically 20 μPa. This is also the default value.
|
/// air, this is typically 20 μPa. This is also the default value.
|
||||||
#[builder(default = "2e-5")]
|
#[builder(default = "Lref_default")]
|
||||||
pub Lref: Flt,
|
pub Lref: Flt,
|
||||||
/// Frequency weightin A/C/Z applied to data. Defaults to [FreqWeighting::default()].
|
/// Frequency weightin A/C/Z applied to data. Defaults to [FreqWeighting::default()].
|
||||||
#[builder(default)]
|
#[builder(default)]
|
||||||
@ -24,6 +28,31 @@ pub struct SLMSettings {
|
|||||||
pub filterDescriptors: Vec<StandardFilterDescriptor>,
|
pub filterDescriptors: Vec<StandardFilterDescriptor>,
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#[cfg_attr(feature = "python-bindings", pymethods)]
|
||||||
|
impl SLMSettings {
|
||||||
|
#[new]
|
||||||
|
#[pyo3(signature=
|
||||||
|
(fs, freqWeighting,
|
||||||
|
timeWeighting,
|
||||||
|
filterDescriptors,
|
||||||
|
Lref=(Lref_default)))]
|
||||||
|
fn new_py(
|
||||||
|
fs: Flt,
|
||||||
|
freqWeighting: FreqWeighting,
|
||||||
|
timeWeighting: TimeWeighting,
|
||||||
|
filterDescriptors: Vec<StandardFilterDescriptor>,
|
||||||
|
Lref: Flt,
|
||||||
|
) -> SLMSettings {
|
||||||
|
SLMSettings {
|
||||||
|
fs,
|
||||||
|
Lref,
|
||||||
|
freqWeighting,
|
||||||
|
timeWeighting,
|
||||||
|
filterDescriptors,
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
#[cfg(test)]
|
#[cfg(test)]
|
||||||
mod test {
|
mod test {
|
||||||
use super::*;
|
use super::*;
|
||||||
@ -31,12 +60,12 @@ mod test {
|
|||||||
|
|
||||||
#[test]
|
#[test]
|
||||||
fn test_slmsettings1() -> Result<()> {
|
fn test_slmsettings1() -> Result<()> {
|
||||||
let desc = StandardFilterDescriptor::genFilterSetInRange(1, 100., 5e3, true).unwrap();
|
let desc = StandardFilterDescriptor::genFilterSetForRange(1, 100., 5e3, true).unwrap();
|
||||||
|
|
||||||
let _ = SLMSettingsBuilder::default()
|
let _ = SLMSettingsBuilder::default()
|
||||||
.fs(1e3)
|
.fs(1e3)
|
||||||
.freqWeighting(FreqWeighting::A)
|
.freqWeighting(FreqWeighting::A)
|
||||||
.timeWeighting(TimeWeighting::Slow)
|
.timeWeighting(TimeWeighting::Slow {})
|
||||||
.filterDescriptors(desc)
|
.filterDescriptors(desc)
|
||||||
.build()
|
.build()
|
||||||
.unwrap();
|
.unwrap();
|
||||||
|
@ -1,3 +1,4 @@
|
|||||||
|
use crate::config::*;
|
||||||
use derive_builder::Builder;
|
use derive_builder::Builder;
|
||||||
use itertools::Itertools;
|
use itertools::Itertools;
|
||||||
use ndarray::ArrayView1;
|
use ndarray::ArrayView1;
|
||||||
@ -22,6 +23,7 @@ struct SLMChannel {
|
|||||||
}
|
}
|
||||||
|
|
||||||
/// Sound Level Meter
|
/// Sound Level Meter
|
||||||
|
#[cfg_attr(feature = "python-bindings", pyclass)]
|
||||||
#[derive(Debug, Clone)]
|
#[derive(Debug, Clone)]
|
||||||
pub struct SLM {
|
pub struct SLM {
|
||||||
// Number of samples processed after last run() is called.
|
// Number of samples processed after last run() is called.
|
||||||
@ -76,16 +78,22 @@ impl SLM {
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
/// Push new time data through sound level meter. Returns L(t) data for each
|
/// Push new time data through sound level meter. Returns L(t) data for each
|
||||||
/// channel.
|
/// channel. Updates the computed statistics and optionally outputs levels
|
||||||
|
/// vs time if flag `provide_output` is set to `true`. Note that at the end
|
||||||
|
/// of the block, the `L(t)` can also be obtained by calling [SLM::Ltlast].
|
||||||
///
|
///
|
||||||
/// # Args
|
/// # Args
|
||||||
///
|
///
|
||||||
/// - `td`: Time data
|
/// - `td`: Time data
|
||||||
pub fn run(&mut self, td: &[Flt]) -> Option<Vec<Vec<Flt>>> {
|
/// - `provide_output` - Set this to true to give intermediate output data
|
||||||
|
pub fn run(&mut self, td: &[Flt], provide_output: bool) -> Option<Vec<Vec<Flt>>> {
|
||||||
if td.len() == 0 {
|
if td.len() == 0 {
|
||||||
return None;
|
return None;
|
||||||
}
|
}
|
||||||
let prefiltered = self.prefilter.filter(td);
|
let prefiltered = self.prefilter.filter(td);
|
||||||
|
|
||||||
|
let level = |a| 10. * Flt::log10(a) / self.Lrefsq;
|
||||||
|
|
||||||
let Lt_iter = self.channels.par_iter_mut().map(|ch| {
|
let Lt_iter = self.channels.par_iter_mut().map(|ch| {
|
||||||
let mut tmp = ch.bp.filter(&prefiltered);
|
let mut tmp = ch.bp.filter(&prefiltered);
|
||||||
let mut N = self.N;
|
let mut N = self.N;
|
||||||
@ -111,21 +119,23 @@ impl SLM {
|
|||||||
// Run filtered_squared signal throug rectifier
|
// Run filtered_squared signal throug rectifier
|
||||||
if let Some(rectifier_down) = &mut ch.rect_lowpass_down {
|
if let Some(rectifier_down) = &mut ch.rect_lowpass_down {
|
||||||
filtered_squared.mapv_inplace(|sample_sq| {
|
filtered_squared.mapv_inplace(|sample_sq| {
|
||||||
|
let rectifier_up = &mut ch.rect_lowpass_up;
|
||||||
|
|
||||||
|
// Asymmetric up/down case for level
|
||||||
let mut fup = sample_sq;
|
let mut fup = sample_sq;
|
||||||
let mut fdown = sample_sq;
|
let mut fdown = sample_sq;
|
||||||
|
|
||||||
// Filter in up-filter
|
// Filter in up-filter
|
||||||
let rectifier_up = &mut ch.rect_lowpass_up;
|
|
||||||
rectifier_up.filter_inout_single(&mut fup);
|
rectifier_up.filter_inout_single(&mut fup);
|
||||||
// Filter in down-filter
|
// Filter in down-filter
|
||||||
rectifier_down.filter_inout_single(&mut fdown);
|
rectifier_down.filter_inout_single(&mut fdown);
|
||||||
|
|
||||||
// Check who wins
|
// Check who wins
|
||||||
if fup > fdown {
|
if fup >= fdown {
|
||||||
rectifier_down.setNextOutputX0(fup);
|
rectifier_down.setToDCValue(fup);
|
||||||
fup
|
fup
|
||||||
} else {
|
} else {
|
||||||
rectifier_up.setNextOutputX0(fdown);
|
rectifier_up.setToDCValue(fup);
|
||||||
fdown
|
fdown
|
||||||
}
|
}
|
||||||
});
|
});
|
||||||
@ -144,11 +154,20 @@ impl SLM {
|
|||||||
});
|
});
|
||||||
// Update last signal power coming from SLM
|
// Update last signal power coming from SLM
|
||||||
ch.stat.Pt_last = *filtered_squared.last().unwrap();
|
ch.stat.Pt_last = *filtered_squared.last().unwrap();
|
||||||
|
// Convert output to levels
|
||||||
|
filtered_squared.mapv_inplace(level);
|
||||||
tmp
|
tmp
|
||||||
});
|
});
|
||||||
|
if provide_output {
|
||||||
let Lt: Vec<_> = Lt_iter.collect();
|
let Lt: Vec<_> = Lt_iter.collect();
|
||||||
self.N += td.len();
|
self.N += td.len();
|
||||||
Some(Lt)
|
Some(Lt)
|
||||||
|
} else {
|
||||||
|
// Just consume the iterator
|
||||||
|
Lt_iter.for_each(|_| {});
|
||||||
|
self.N += td.len();
|
||||||
|
None
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Number of channels in SLM
|
/// Number of channels in SLM
|
||||||
@ -163,7 +182,7 @@ impl SLM {
|
|||||||
Dcol::from_iter(
|
Dcol::from_iter(
|
||||||
self.channels
|
self.channels
|
||||||
.iter()
|
.iter()
|
||||||
.map(|ch| 20. * Flt::log10(stat_returner(ch) / self.Lrefsq)),
|
.map(|ch| 10. * Flt::log10(stat_returner(ch) / self.Lrefsq)),
|
||||||
)
|
)
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -186,6 +205,36 @@ impl SLM {
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#[cfg_attr(feature = "python-bindings", pymethods)]
|
||||||
|
impl SLM {
|
||||||
|
#[new]
|
||||||
|
fn new_py(settings: SLMSettings) -> SLM {
|
||||||
|
SLM::new(settings)
|
||||||
|
}
|
||||||
|
|
||||||
|
#[pyo3(name = "run", signature=(dat, provide_output=true))]
|
||||||
|
fn run_py(&mut self, dat: PyArrayLike1<Flt>, provide_output: bool) -> Option<Vec<Vec<Flt>>> {
|
||||||
|
self.run(dat.as_array().as_slice()?, provide_output)
|
||||||
|
}
|
||||||
|
|
||||||
|
#[pyo3(name = "Lmax")]
|
||||||
|
fn Lmax_py<'py>(&self, py: Python<'py>) -> PyArr1Flt<'py> {
|
||||||
|
PyArray1::from_array_bound(py, &self.Lmax())
|
||||||
|
}
|
||||||
|
#[pyo3(name = "Leq")]
|
||||||
|
fn Leq_py<'py>(&self, py: Python<'py>) -> PyArr1Flt<'py> {
|
||||||
|
PyArray1::from_array_bound(py, &self.Leq())
|
||||||
|
}
|
||||||
|
#[pyo3(name = "Lpk")]
|
||||||
|
fn Lpk_py<'py>(&self, py: Python<'py>) -> PyArr1Flt<'py> {
|
||||||
|
PyArray1::from_array_bound(py, &self.Lpk())
|
||||||
|
}
|
||||||
|
#[pyo3(name = "Ltlast")]
|
||||||
|
fn Ltlast_py<'py>(&self, py: Python<'py>) -> PyArr1Flt<'py> {
|
||||||
|
PyArray1::from_array_bound(py, &self.Ltlast())
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
#[derive(Debug, Clone, Default)]
|
#[derive(Debug, Clone, Default)]
|
||||||
/// Quantities defined as powers, i.e. square of amplitude
|
/// Quantities defined as powers, i.e. square of amplitude
|
||||||
struct SLMStat {
|
struct SLMStat {
|
||||||
@ -213,13 +262,13 @@ mod test {
|
|||||||
#[test]
|
#[test]
|
||||||
fn test_slm1() {
|
fn test_slm1() {
|
||||||
const fs: Flt = 48e3;
|
const fs: Flt = 48e3;
|
||||||
const N: usize = (fs/8.) as usize;
|
const N: usize = (fs / 8.) as usize;
|
||||||
|
|
||||||
let desc = StandardFilterDescriptor::Overall().unwrap();
|
let desc = StandardFilterDescriptor::Overall().unwrap();
|
||||||
|
|
||||||
let settings = SLMSettingsBuilder::default()
|
let settings = SLMSettingsBuilder::default()
|
||||||
.fs(fs)
|
.fs(fs)
|
||||||
.timeWeighting(TimeWeighting::Fast)
|
.timeWeighting(TimeWeighting::Fast {})
|
||||||
.freqWeighting(FreqWeighting::Z)
|
.freqWeighting(FreqWeighting::Z)
|
||||||
.filterDescriptors(&[desc])
|
.filterDescriptors(&[desc])
|
||||||
.build()
|
.build()
|
||||||
@ -230,14 +279,12 @@ mod test {
|
|||||||
siggen.reset(fs);
|
siggen.reset(fs);
|
||||||
let mut data = vec![0.; N];
|
let mut data = vec![0.; N];
|
||||||
siggen.genSignal(&mut data);
|
siggen.genSignal(&mut data);
|
||||||
// println!("{:#?}", data);
|
|
||||||
|
|
||||||
let mut slm = SLM::new(settings);
|
let mut slm = SLM::new(settings);
|
||||||
// println!("{slm:#?}");
|
// println!("{slm:#?}");
|
||||||
let res = slm.run(&data).unwrap();
|
let res = slm.run(&data, true).unwrap();
|
||||||
let res = &res[0];
|
let res = &res[0];
|
||||||
println!("{slm:#?}");
|
println!("{slm:#?}");
|
||||||
println!("{:#?}", &res[res.len()- 100..]);
|
println!("{:#?}", &res[res.len() - 100..]);
|
||||||
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -1,18 +1,23 @@
|
|||||||
use crate::Flt;
|
use crate::config::*;
|
||||||
/// Time weighting to use in level detection of Sound Level Meter.
|
/// Time weighting to use in level detection of Sound Level Meter.
|
||||||
#[derive(Clone, Copy)]
|
#[cfg_attr(feature = "python-bindings", pyclass(eq))]
|
||||||
|
#[derive(Clone, Copy, PartialEq)]
|
||||||
pub enum TimeWeighting {
|
pub enum TimeWeighting {
|
||||||
|
// I know that the curly braces here are not required and add some
|
||||||
|
// boilerplate, but this is the only way Pyo3 swallows complex enums at the
|
||||||
|
// moment.
|
||||||
/// Slow time weighting ~ 1 s
|
/// Slow time weighting ~ 1 s
|
||||||
Slow,
|
Slow {},
|
||||||
/// Fast time weighting ~ 1/8 s
|
/// Fast time weighting ~ 1/8 s
|
||||||
Fast,
|
Fast {},
|
||||||
/// Impulse time weighting ~ 30 ms
|
/// Impulse time weighting ~ 30 ms
|
||||||
Impulse,
|
Impulse {},
|
||||||
/// A custom symmetric time weighting
|
/// A custom symmetric time weighting
|
||||||
CustomSymmetric {
|
CustomSymmetric {
|
||||||
/// Custom time constant [s]
|
/// Custom time constant [s]
|
||||||
t: Flt,
|
t: Flt,
|
||||||
},
|
},
|
||||||
|
|
||||||
/// A custom symmetric time weighting
|
/// A custom symmetric time weighting
|
||||||
CustomAsymmetric {
|
CustomAsymmetric {
|
||||||
/// Time weighting when level is increasing
|
/// Time weighting when level is increasing
|
||||||
@ -23,7 +28,7 @@ pub enum TimeWeighting {
|
|||||||
}
|
}
|
||||||
impl Default for TimeWeighting {
|
impl Default for TimeWeighting {
|
||||||
fn default() -> Self {
|
fn default() -> Self {
|
||||||
TimeWeighting::Fast
|
TimeWeighting::Fast {}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
impl TimeWeighting {
|
impl TimeWeighting {
|
||||||
@ -32,13 +37,13 @@ impl TimeWeighting {
|
|||||||
pub fn getLowpassPoles(&self) -> (Flt, Option<Flt>) {
|
pub fn getLowpassPoles(&self) -> (Flt, Option<Flt>) {
|
||||||
use TimeWeighting::*;
|
use TimeWeighting::*;
|
||||||
match self {
|
match self {
|
||||||
Slow => (-1.0, None),
|
Slow {} => (-1.0, None),
|
||||||
Fast =>
|
Fast {} =>
|
||||||
// Time constant is 1/8 s, pole is at -8 rad/s
|
// Time constant is 1/8 s, pole is at -8 rad/s
|
||||||
{
|
{
|
||||||
(-8., None)
|
(-8., None)
|
||||||
}
|
}
|
||||||
Impulse => {
|
Impulse {} => {
|
||||||
// For the impulse time weighting, some source says ~ 2.9 dB/s
|
// For the impulse time weighting, some source says ~ 2.9 dB/s
|
||||||
// drop for the decay
|
// drop for the decay
|
||||||
// [https://www.nti-audio.com/en/support/know-how/fast-slow-impulse-time-weighting-what-do-they-mean].
|
// [https://www.nti-audio.com/en/support/know-how/fast-slow-impulse-time-weighting-what-do-they-mean].
|
||||||
|
@ -1,62 +0,0 @@
|
|||||||
#!/usr/bin/env python3
|
|
||||||
import numpy as np
|
|
||||||
from lasprs.filter import Biquad, SeriesBiquad, BiquadBank
|
|
||||||
|
|
||||||
|
|
||||||
def test_biquad1():
|
|
||||||
"""
|
|
||||||
Test if manual unit biquad matches response for manual input biquad.
|
|
||||||
"""
|
|
||||||
input_ = np.array([1.0, 0, 0, 0, 0, 0, 0])
|
|
||||||
|
|
||||||
b = Biquad(np.array([1., 0, 0, 1, 0, 0]))
|
|
||||||
b2 = Biquad.unit()
|
|
||||||
out1 = b.filter(input_)
|
|
||||||
out2 = b2.filter(input_)
|
|
||||||
|
|
||||||
expected_output = input_
|
|
||||||
assert(np.linalg.norm(out1 - out2) == 0.)
|
|
||||||
assert(np.linalg.norm(out1 - expected_output) == 0.)
|
|
||||||
|
|
||||||
def test_seriesbiquad():
|
|
||||||
"""
|
|
||||||
Test if manual unit biquad matches response for manual input biquad.
|
|
||||||
"""
|
|
||||||
input_ = np.array([1.0, 0, 0, 0, 0, 0, 0])
|
|
||||||
|
|
||||||
f1 = [1., 0, 0, 1, 0, 0]
|
|
||||||
f2 = [1., 0, 0, 1, 0, 0]
|
|
||||||
f = f1+f2
|
|
||||||
|
|
||||||
# Two biquads in series
|
|
||||||
b = SeriesBiquad(np.array(f))
|
|
||||||
|
|
||||||
# Single one
|
|
||||||
b2 = SeriesBiquad.unit()
|
|
||||||
|
|
||||||
out1 = b.filter(input_)
|
|
||||||
out2 = b2.filter(input_)
|
|
||||||
|
|
||||||
expected_output = input_
|
|
||||||
assert(np.linalg.norm(out1 - out2) == 0.)
|
|
||||||
assert(np.linalg.norm(out1 - expected_output) == 0.)
|
|
||||||
|
|
||||||
def test_biquadbank():
|
|
||||||
"""
|
|
||||||
See if two filters with half gain produce the output=input
|
|
||||||
"""
|
|
||||||
input_ = np.array([1.0, 0, 0, 0, 0, 0, 0])
|
|
||||||
f1 = [1., 0, 0, 1, 0, 0]
|
|
||||||
f2 = [1., 0, 0, 1, 0, 0]
|
|
||||||
filters = np.array([f1, f2]).T
|
|
||||||
bank = BiquadBank(filters)
|
|
||||||
bank.set_gains([0.5, 0.5])
|
|
||||||
filtered = bank.filter(input_)
|
|
||||||
assert(bank.len() == 2)
|
|
||||||
assert(np.linalg.norm(input_ - filtered) == 0)
|
|
||||||
|
|
||||||
|
|
||||||
if __name__ == '__main__':
|
|
||||||
test_biquad1()
|
|
||||||
test_seriesbiquad()
|
|
||||||
test_biquadbank()
|
|
Loading…
Reference in New Issue
Block a user