Someone in the physics stack exchange told me to come here for this question. My question is how does a person's voice affect how a soundwave looks? For example, if I say the letter "a" it would sound different than if you said the letter "a", but my question is if there is a pattern to this, or maybe if we are able to derive a waveform based on voice (however it would be affected by multiple other factors due to when I say "a" it is different than when I say "b").
$\begingroup$
$\endgroup$
Add a comment
|
3 Answers
$\begingroup$
$\endgroup$
2
I would view it as the following:
The letter A is the input to a system, one is your vocal chords the other one mine. Since the system response of your and my vocal chords are different, we would get different outputs.
Now, what would characterize the system response, it would be factors like, rise time/bandwidth, decay response etc. All of this would impact the tone and pitch terminologies usually used in audio/speech. These variables will vary for each user. With different input the response will then also be different . So "b" will not be same as "a"
-
$\begingroup$ Yes, so my question is whether it is possible/ how to get the function that turns that input into an output for a certain voice. Or perhaps to create a default "voice" that we can get a function to transform it into a certain voice. $\endgroup$– M SCommented Jun 19, 2020 at 23:57
-
$\begingroup$ The research of Stylianou allowed for turning the voice of a young woman into eg an elderly woman: cs.cmu.edu/~pmuthuku/mlsp_page/lectures/Stylianou_VC.pdf $\endgroup$ Commented Jun 20, 2020 at 6:14
$\begingroup$
$\endgroup$
Someone wrote that spectrograms were introduced to help deaf and hearing-impaired to "read" speech. Unfortunately, it is quite hard to interpret what is being said from a spectrogram, but I am told that a few people (working in speech recognition?) are able to.
Depending on the speech source, the recording quality and the parameters of the spectrogram, you should be able to clearly identify "pitched" sounds ("a", "z") by the comb-like horizontal structure of harmonics. A male speaker would typically have a deeper voice, apparent as a shift towards lower frequencies. Fricatives like "s" and "t" will appear like wide-band, unstructured "noise" with some temporal and frequency limits.
It is obvious that there is some general pattern transcending the individual: I am able to listen to you talking and comprehend what you are saying, and so is automatic speech recognition. I do believe that I am more robust with talkers that I know, or talkers that belong to "my" kind of dialect. Thus individual deviation does matter.
$\begingroup$
$\endgroup$
Well in one sense Voice is the soundwave.
But I think you are asking about accent or individuals pronunciation vs the the idealized sound abstracted from accent or specific pronunciation.
So you could just compare two voices, or try to compare one voice to an abstract ideal. Even for saying "A".
I think the signal of speech is so complicated your question can't be answered.
