Build a Vocoder Using STFT Transformations

Impose the spectral envelope of a modulator signal onto a carrier

Open in Cloud
Download Notebook

Define the modulator signal using

SpeechSynthesize

In[1]:=

voice=AudioTimeStretch[SpeechSynthesize["we are the robots"],1.4]

Out[1]=

00:00

00:02

Data in Notebook

Also define the carrier:

In[2]:=

carrier=Audio@Sound[{SoundNote[{"C2","C3","G","C5"},.5*1.4,"BrassSection"],SoundNote[{"D#2","D#3","ASharp","D#5"},.3*1.4,"BrassSection"],SoundNote[{"C2","C3","G","C5"},1,"BrassSection"]}]

Out[2]=

00:00

00:02

Data in Notebook

Make sure that the carrier has enough spectral content to be filtered by the modulator:

In[3]:=

{voice,carrier}=ConformAudio[{voice,carrier}];

Using

MeanFilter

, extract the spectral envelope of the magnitude spectrogram:

In[4]:=

stftVoice=ShortTimeFourier[voice,2048,512,HannWindow];modulator=Rescale[MeanFilter[Abs@#,20]&/@stftVoice["Data"]];

Now modify the short-time Fourier transform of the carrier using the modulator:

In[5]:=

stftCarrier=ShortTimeFourier[carrier,2048,512,HannWindow];stftCarrier["Data"]=modulator*stftCarrier["Data"];

Then reconstruct the audio using

InverseShortTimeFourier

In[6]:=

InverseShortTimeFourier[stftCarrier]//AudioNormalize

Out[6]=

00:00

00:02

Data in Notebook

Related Symbols

SpeechSynthesize
ShortTimeFourier
InverseShortTimeFourier
AudioTimeStretch
AudioNormalize
Audio
SoundNote
Sound

Publisher Information

Contributed by: Wolfram Staff