Wolfram Audio | Things to Try

Make edits and run any piece of code by clicking inside the code and pressing
+
.
Audio Processing & Analysis. Wolfram Audio offers highly optimized processing and high-level analysis of speech, music and other audio signals. Tight integration with machine learning and neural networks enables solutions in automated systems, security, medicine and more.

Import, Process and Export Audio

Import an audio file with high-frequency background noise:
Run
In[]:=
music=Import["ExampleData/sample.flac"]
Show the audio spectrogram:
Run
In[]:=
Spectrogram[music]
Remove the high-frequency background noise:
Run
In[]:=
filteredmusic=LowpassFilter[music,
800
Hz
]
See the spectrogram of the filtered audio:
Run
In[]:=
Spectrogram[filteredmusic]
Export the result in the format of your choice:
Run
In[]:=
Export["filteredexample.mp3",filteredmusic]

Visualize Audio Signals

Audio from different types of sources:
Run
In[]:=
sources=ExampleData[{"Audio",#}]&/@{"Cat","Cello","FemaleVoice"}
Audio waveforms show audio amplitude over time:
Run
In[]:=
AudioPlot[#,PlotLayout->"Averaged"]&/@sources
Spectrograms show frequency over time:
Run
In[]:=
Spectrogram/@sources
Periodograms show dominant frequencies:
Run
In[]:=
Periodogram[#,256]&/@sources

Remove Noise or Add Effects

Start with a noisy audio clip:
Run
In[]:=
apollo=ExampleData[{"Audio","Apollo11SmallStep"}]
Denoise the audio:
Run
In[]:=
filteredapollo=WienerFilter[apollo,30]
Perform a pitch shifting on the first half of the audio:
Run
In[]:=
{beginning,end}=AudioSplit[filteredapollo,4];​​AudioJoinAudioPitchShiftbeginning,
1
octaves
,end

Extract Features from Audio

Begin with a dataset of spoken digits:
Run
In[]:=
digitspeech=
Input
SpeakerID
Output
Speaker A
1
Speaker A
2
Speaker A
3
Speaker B
1
Speaker B
2
Speaker B
3
Speaker C
1
Speaker C
2
Speaker C
3
;
Define a feature extractor using a pre-trained model:
Run
In[]:=
extractor=NetAppend[NetTake[NetModel["Wav2Vec2 Trained on LibriSpeech Data"],"FeatureExtractor"],"Mean"->AggregationLayer[Mean,1]]
Show extracted features in a 3D plot:
Run
In[]:=
Module{colors,styling},​​colors="Speaker A"->
,"Speaker B"->
,"Speaker C"->
;​​styling=
Function[
]
;​​Legended[​​FeatureSpacePlot3D[Normal[digitspeech[All,styling]],FeatureExtractor->extractor,LabelingFunction->None],​​PointLegend[Values[colors],Keys[colors]]​​]​​

Identify Audio Sources

Identify what’s in an audio signal:
Run
In[]:=
AudioIdentify

Begin with audio from mixed sources:
Run
In[]:=
mixedsources=
;
Identify the dominant source by half-second segments:
Run
In[]:=
segmentIds=AudioBlockMapAudioIdentify
&,mixedsources,{1,.5}//Normal
Assemble the segments into intervals:
Run
In[]:=
intervals=GroupBysegmentIds,Last,
Function[
]

Plot the results:
Run
In[]:=
LegendedAudioPlotmixedsources,

,SwatchLegend
