Machine learning techniques in speech recognition:

[creative64]

Hi,

I'd like to know how are Machine learning techniques being used in modern speech recognition
systems ?

Few of the things that come to mind are:

1) Language modelling: Replace LMs with say neural nets.

2) Intent processing: Interpret the intent of spoken speech. For example one car driver might say
"It's very cold in here, can you please increase the Cooling" or a different
driver might say "Hey its very cold, jack-up the cooling" and so on.
Both these statements have the same intended action and that is increasing the
cooling.

3) Language translation: Interpreted hypothesis is translated into another language.

My questions are:

a) Are machine learning techniques actually used to solve any of the problems mentioned above ?

b) Any other areas where ML techniques could be used in speech recognition ?

c) Any recommended good material on the subject?

Thanks and regards,

[Nickolay V. Shmyrev]

Machine learning techniques being used in modern speech recognition

Speech recognition is a subdomain of machine learning by itself

1) Language modelling: Replace LMs with say neural nets.

This is successfully done with recursive neural networks. You can check rnnlm toolkit

2) Intent processing: Interpret the intent of spoken speech. For example one car driver might say

This can be done with opennlp

a) Are machine learning techniques actually used to solve any of the problems mentioned above ?

Yes

b) Any other areas where ML techniques could be used in speech recognition ?

Like I wrote above there are way more things, for example G2P learning is a classical machine learning task too.

c) Any recommended good material on the subject?

You'd better focus on something specific.