Voice identification!
How much experience do you have with ML and DL? The way you phrased the question in the title sounds like a “callback” to the classic “feature engineering” approach that people used from the beginning of the computer era through maybe the late 1990s, before ML based on Neural Networks really flowered. ML has changed everything. Now you don’t have to figure out what the features are: you just create a neural network of the appropriate style and complexity, define your goals in terms of objective functions and then throw a massive amount of labelled training data at the network and let it figure out what the features are by learning through back propagation.
Voice recognition is a very well developed field at this point and there is a lot of literature out there about how to approach this and lots of existing products that you can use. So it depends on what your goals are here:
Are you trying to build an actual application and does it work for you to use someone else’s system as the basis and just put your application specific UI in front of it?
Or is this some kind of academic project and you have to build a system yourself to demonstrate that you understand the concepts?
Obviously those are totally different scenarios that are the endpoints of a large span and there are other points between those extremes.
If you’d like to get some sense of what commercial products are out there, just googling “voice recognition” will get you started. If you want to know more about the academic state of the art, start with Papers with Code and search for voice.
1 Like
Sir I am just a beginner, 20 yrs old engineering student . Training a neural network on voice data would work but human voice changes over age so it won’t be a perfect solution. What I am trying to do is extract features store them ,then when at time of identification take users voice extract features and compare them with stored one and if it matches over a certain threshold allow access, then replace the stored with new ones for each sucessful identification for a single user. (Features would be extracted for a fixed phrase as “Hello my name is Vedant”)
Well, if your voice can change with age, then why can’t the features change also if you are thinking of this in terms of features?
I would say changes to the inputs over time is an orthogonal problem. Your face changes over time also, but are there enough features that are retained that a face recognition algorithm would continue to work? It’s an interesting question and I don’t claim to know the answer, but I’ll bet we can find some papers about that. 
My suggestion would be first to solve the base problem without considering the “evolution over time” issue. Then once you have a valid solution to the base problem, you can consider how to evaluate the behavior of your model over time. But this will only be possible if your training dataset has samples of given people’s voices that are also over time.
Your particular application sounds exactly analogous to one of the topics about Face Recognition that is covered here by Prof Andrew Ng in the Deep Learning Specialization Course 4 about Convolutional Nets. He has a section in Week 4 about Face Recognition that covers both the “one shot” case (“is this person at the door someone in my database”) and the general problem (“do these two pictures represent the same person even if it’s not someone from my database”). If you are just getting started on ML/DL, one good approach would be to take some of the courses offered here. If you’re starting from scratch, the normal path would be to take the Machine Learning Specialization first and then do the Deep Learning Specialization.
2 Likes
Thankyou for the guidance sir , I am currently doing the Machine learning specialization and then will continue with deep learning specialization. Also can you suggest website where I can find research papers and about people’s work regarding machine learning topics .
You will find an internet search to be very useful. There is no single website that provides all possible links to research topics.
Check the Papers with Code link that I gave you earlier, but as Tom says, the Internet is a big place and, as they say, “Google is your friend!” 