How do you remove noise to detect just the human voice?

How do you detect different sound frequencies and cut off parts in an audio file? Among instruments, how do you pick up the human voice?

• Wasn't this question just posted and migrated to Signal Processing? Nov 27, 2012 at 3:50
• @s0rce could you post the URL ? Nov 27, 2012 at 4:19
• @VitaliyKaurov This was the original question. I migrated it, but the OP deleted it and reasked it here again...
– rm -rf
Nov 27, 2012 at 5:06
• Welcome to Mathematica.SE! I suggest the following: 1) As you receive help, try to give it too, by answering questions in your area of expertise. 2) Read the FAQs! 3) When you see good Q&A, vote them up by clicking the gray triangles, because the credibility of the system is based on the reputation gained by users sharing their knowledge. ALSO, remember to accept the answer, if any, that solves your problem, by clicking the checkmark sign Nov 27, 2012 at 6:32
• Do you know what type of noise has been added, periodic, narrow band, wide band; white, pink etc? Nov 27, 2012 at 13:58

A lot depends on your specific data. But if the noise is far from voice in frequency domain there is a simple brute-force trick of cutting off/out "bad" frequencies using wavelets. Let's import some sample recording:

voice = ExampleData[{"Sound", "Apollo11ReturnSafely"}]


WaveletScalogram is great for visualizing voice versus noise features:

cwt = ContinuousWaveletTransform[voice, GaborWavelet[6]];
WaveletScalogram[cwt, ColorFunction -> "AvocadoColors", ColorFunctionScaling -> False]


Voice is more rich and irregular in structure, noise is more monotonic and repetitive. So now based on the visual we can formulate a logical condition to cut out the noisy octaves (numbers on vertical axes):

cwtCUT = WaveletMapIndexed[#1 0.0 &, cwt, {u_ /; u >= 6 && u < 9, _}];
WaveletScalogram[cwtCUT, ColorFunction ->"AvocadoColors", ColorFunctionScaling -> False]


This is pretty brutal, like a surgery that cuts out good stuff too, because in this cases some voice frequencies blend with noise and we lost them. But it roughly works - signal is cleaner. You can hear how many background noises were suppressed (a few still stay though) - use headphones or good speakers. If in your cases noise is even further from voice in frequency domain - it will work much better.

InverseContinuousWaveletTransform[cwtCUT]


• How did you make the decision on what was and wasn't noise? Nov 27, 2012 at 5:44
• @rcollyer I gave some explanation under WaveletScalogram image. Voice has rich irregular structure. Noise is monotonic and repetitive. Nov 27, 2012 at 5:47
• @NasserM.Abbasi Thanks! I corrected the code, should work now. Nov 27, 2012 at 6:20
• Maybe I should have read it ... :P Nov 27, 2012 at 6:24

What you need is BandpassFilter, which is new in version 9. Assuming your audio is sampled at 22400 Hz, you can do:

BandpassFilter[data, {60 π, 180 π}, SampleRate -> 22400]


to filter it to between 60-180 Hz.

• This works for SampleSoundList, so for voice above, this would be done: Sound[BandpassFilter[voice[[1]], {60 π, 180 π}, SampleRate -> 22400]]. It did not work as well as the wavelet based denoising by @Vitaliy Kaurov Dec 2, 2012 at 10:03
• @ubpdqn You'll have to filter the correct band for the sound sample that Vitaliy used. I was addressing OP's question where they wanted to filter their data (not shared) between 60 and 180 Hz.
– rm -rf
Dec 2, 2012 at 15:23
• @rm-rf thank you for the clarification Dec 3, 2012 at 8:30
• Should that be BandpassFilter[data, {2 π 60, 2 π 180}, SampleRate -> 22400] for radian frequencies? That is, a coefficient of 2 π rather than π. Dec 3, 2012 at 18:06
• @david Maybe... I wrote this answer before version 9 was actually released (going just by the docs which rolled out early). I'll test it with an example edit it later today with some more info.
– rm -rf
Dec 3, 2012 at 18:08

About a year ago,I saw a demo in Labview that can detect the voice of killer whale in a setting of the sound of seawater.

I want to try the similar thing in Mathematica. Based upon Vitaliy Kaurov's approach:

voice = ExampleData[{"Sound", "Apollo11ReturnSafely"}];
data = voice[[1, 1, 1]]; r = voice[[1, 2]];
cwt = ContinuousWaveletTransform[data,
GaborWavelet[6]];(*If you set cwt=ContinuousWaveletTransform[data,GaborWavelet[6],{Automatic,8}];
you will get more accurate result.But you must re-extract the interest region*)
WaveletScalogram[cwt, ColorFunction -> "AvocadoColors", ColorFunctionScaling -> False]


It gives the the scalogram of the wave.

Then you can use mma graph tools to describe your outline.

Firstly, press Ctrl+D to open the graph tools.

Secondly, press the button in the lower right corner.

Thirdly, you can get the coordinates(use Ctrl+C and Ctrl+V).

My data is the following result

test = {{7183, 40.14}, {7309, 39.89}, {7771, 39.77}, {7939, 39.64}, {8065, 39.39}, {8863, 38.64}, {9913, 37.15}, {1.067*^4, 35.9}, {1.096*^4, 35.65}, {1.13*^4, 35.27}, {1.163*^4, 35.15}, {1.201*^4, 35.02}, {1.247*^4, 35.02}, {1.306*^4, 35.02}, {1.369*^4, 35.27}, {1.428*^4, 35.65}, {1.47*^4, 35.9}, {1.495*^4, 36.15}, {1.52*^4, 36.4}, {1.541*^4, 36.77}, {1.587*^4, 37.27}, {1.629*^4, 37.64}, {1.671*^4, 38.02}, {1.726*^4, 38.14}, {1.789*^4, 38.27}, {1.873*^4, 38.27}, {1.957*^4, 38.27}, {2.02*^4, 38.02}, {2.062*^4, 38.02}, {2.083*^4, 38.02}, {2.104*^4, 38.14}, {2.129*^4, 38.14}, {2.184*^4, 38.14}, {2.238*^4, 37.77}, {2.276*^4, 37.39}, {2.314*^4, 37.02}, {2.343*^4, 36.65}, {2.373*^4, 36.4}, {2.41*^4, 35.77}, {2.427*^4, 35.52}, {2.431*^4, 35.02}, {2.415*^4, 33.53}, {2.402*^4, 32.9}, {2.373*^4, 32.53}, {2.322*^4, 32.28}, {2.314*^4, 32.15}, {2.259*^4, 32.15}, {2.217*^4, 32.15}, {2.196*^4, 32.03}, {2.179*^4, 32.15}, {2.163*^4, 31.91}, {2.121*^4, 31.53}, {2.104*^4,31.16}, {2.066*^4, 30.91}, {2.033*^4, 30.66}, {1.999*^4, 30.53}, {1.961*^4, 30.53}, {1.911*^4, 30.53}, {1.852*^4, 30.53}, {1.81*^4, 30.53}, {1.751*^4, 30.66}, {1.709*^4, 30.91}, {1.671*^4, 30.91}, {1.646*^4, 31.03}, {1.625*^4, 31.03}, {1.6*^4, 31.03}, {1.579*^4, 30.91}, {1.562*^4, 30.66}, {1.537*^4, 30.16}, {1.512*^4, 29.91}, {1.495*^4, 29.66}, {1.478*^4, 29.53}, {1.449*^4, 29.41}, {1.394*^4, 29.28}, {1.357*^4, 29.28}, {1.31*^4, 29.41}, {1.256*^4, 29.66}, {1.214*^4, 29.78}, {1.176*^4, 29.78}, {1.142*^4, 29.78}, {1.088*^4, 29.78}, {1.046*^4, 29.91}, {1.021*^4, 29.91}, {9913, 29.78}, {9745, 29.53}, {9493, 29.28}, {9241, 28.91}, {8947, 28.66}, {8737, 28.29}, {8527, 27.91}, {8317,     27.54}, {8065, 27.16}, {7855, 26.66}, {7603, 26.04}, {7267, 25.54}, {6973, 24.92}, {6806, 24.42}, {6554, 24.04}, {6344, 23.67}, {6176, 23.42}, {6050, 23.3}, {5840, 23.17}, {5714, 23.17}, {5672, 23.67}, {5588, 24.42}, {5546, 25.29}, {5546,     26.54}, {5546, 27.66}, {5546, 28.41}, {5546, 29.16}, {5588, 30.16}, {5630, 31.16}, {5630, 31.91}, {5672, 32.78}, {5672, 33.53}, {5672, 34.15}, {5672, 34.77}, {5672, 35.52}, {5714, 36.27}, {5714, 36.77}, {5714, 37.39}, {5714, 37.77}, {5756, 38.02}, {5756, 38.39}, {5840, 38.77}, {6008, 39.39}, {6092,     39.52}, {6176, 39.64}, {6302, 39.89}, {6428, 39.89}, {6554, 40.14}, {6638, 40.14}, {6764, 40.14}, {6848, 40.14}};
ListPlot@test


Then I define a function to transform the coordinates to the coordinates in WaveletScalogram:

g[{x_, y_}] :=
Module[{a =
Floor[(cwt["Octaves"] + 1) - y/cwt["Voices"],
1./cwt["Voices"]]}, {x, {Floor[a],
Floor[(a - Floor[a])*cwt["Voices"]] + 1}}];


In addition, I define a function to smoothen the coordinates:

smooth[lis_] := 1/3*(Total /@ Partition[RotateRight@lis, 3, 1, 1])


And

smoothtestdata = smooth@test; {ymin, ymax} =
Through[{Ceiling@Min@# &, Floor@Max@# &}[smoothtestdata[[All, 2]]]];
WaveletCoordinate = g /@ (Round@
Module[{gra},
gra = ListLinePlot[Append[smoothtestdata, smoothtestdata[[1]]],
MeshFunctions -> Function[{x, y}, y],
Mesh -> {Range[ymin, ymax, 1]}];
Cases[Normal@gra, Point[ptlist_] :> ptlist, Infinity] //
SortBy[#, Last] &])


I get the result:

{{5638, {8, 1}}, {6541, {8, 1}}, {7035, {7, 4}}, {5578, {7, 4}}, {5552, {7, 3}}, {7534, {7, 3}}, {5546, {7, 2}}, {8022, {7, 2}}, {8576, {7, 1}}, {5546, {7, 1}}, {5556, {6, 4}}, {9273, {6, 4}}, {5583, {6, 3}}, {15207, {6, 3}}, {16069, {6, 2}}, {16414, {6, 2}}, {20768, {6, 2}}, {5614, {6, 2}}, {5645, {6, 1}}, {21748, {6, 1}}, {23970, {5, 4}}, {5663, {5, 4}}, {24177, {5, 3}}, {5672, {5, 3}}, {5676, {5, 2}}, {24236, {5, 2}}, {5696, {5, 1}}, {23957, {5, 1}}, {14766, {5, 1}}, {10686, {5, 1}}, {5714, {4, 4}}, {15657, {4, 4}}, {23130, {4, 4}}, {9977, {4, 4}}, {9241, {4, 3}}, {5739, {4, 3}}, {16923, {4, 3}}, {21869, {4, 3}}, {8466, {4, 2}}, {5913, {4, 2}}, {6464, {4, 1}}, {7255, {4, 1}}}


The first element of each sublist is time (Surely, I have not considered the SampleRate now), the second is coordinate in wavelet ({Octaves,Voices})

In order to detect the interest region,I define a function.

f[lis_, pos_] :=
Module[{poslen = Length@pos, temp},
temp = ReplacePart[lis, i_ /; (i < pos[[1]]) -> 0];
Do[temp = ReplacePart[temp,
i_ /; (pos[[index]] < i < pos[[index + 1]]) -> 0], {index, 2, poslen - 2, 2}];
temp = ReplacePart[temp, i_ /; (i > pos[[poslen]]) -> 0]; temp]


Finally, set the irrelevant region to zero:

Module[{temp, tempwavelet = cwt},
Do[temp = Transpose[GatherBy[WaveletCoordinate, Last][[i]]];
tempwavelet = WaveletMapIndexed[f[#, Sort@temp[[1]]] &, tempwavelet,
temp[[2, 1]]],
{i, 1, Length@GatherBy[WaveletCoordinate, Last]}];
tempwavelet =
WaveletMapIndexed[0.*# &, tempwavelet,
Except[Alternatives @@ WaveletCoordinate[[All, 2]]]]; tempwavelet]
WaveletScalogram[%, ColorFunction -> "AvocadoColors",  ColorFunctionScaling -> False]


If we transform the data to sound:

SampledSoundList[InverseContinuousWaveletTransform[%%], r] // Sound


we get:

You can hear the human voice more clearly! :)

About the function f: Think about the following graph:

If the orange region is data region,the green region is my interest,I want to extract the octave 2 and voice 1,I can use the code:

f[Range[10], {1, 6}](*Because my interest time is 1-6*)
(*result: {1,2,3,4,5,6,0,0,0,0}*)


Extract the octave 3 and voice 3:

f[Range[10], {4, 8}](*Because my interest time is 4-8*)
(*result: {0,0,0,4,5,6,7,8,0,0}*)


Extract the octave 3 and voice 4:

f[Range[10], {5,7}](*Because my interest time is 5-7*)
(*result: {0,0,0,0,5,6,7,0,0,0}*)
`

So If combine this function and WaveletMapIndexed,we can extract the data.

Let me guess,If we don't extract by hand,otherwise,use the picture processing to get the outline and remove the noise color from the voice color,What's like?

• I tried to improve the format and expression, but since I'm not familiar with this issue, what I can do is quite limited and I might have made mistakes, so feel free to roll back if you don't like my edit. Jan 7, 2016 at 6:09
• Thank you.more beautiful than ever. Jan 7, 2016 at 6:42