Emission probability matrix in Hidden Markov Model Can an expert let me help in understanding the meaning of Response parameters. This is the result of running HMM from Depmixs4 package of R. St, st2 are states. 

If it's in Gaussian distribution PDF then what is intercept and sd here? How would i calculate probability from it?
 A: You can find an account of the models and fitting methods used in the depmixS4 package here.  As you can see from the documentation, the hidden Markov model consists of a series of hidden state variables that constitute a Markov chain, and a series of observable random variables that have a specified distribution with parameters that differ depending on the underlying hidden state variable.  The documentation for the package specifies that the parameters in the model are fit by maximum likelihood estimation, by iterative use of the EM algorithm or a Newton-Raphson optimiser.

In your case you have a two-state model where the output in each state comes from a Gaussian distributions with unknown mean and variance, and there is a transition matrix with transition probabilities between these two states.  Your model consists of a series of hidden (binary) state variables $S_0, S_1, S_2, ... $ which follow a Markov chain with transition matrix:
$$\mathbf{P} = \begin{bmatrix} 
\theta & 1-\theta \\[6pt]
1-\phi & \phi \\[6pt]
\end{bmatrix},$$
where $\theta \equiv \mathbb{P}(S_{t+1}=0 | S_t=0)$ and $\phi \equiv  \mathbb{P}(S_{t+1}=1 | S_t=1)$ are the unknown parameters determining the probability of remaining in the existing state.  You then have a series of observable variables which follow the state-conditional distributions:
$$X_t | S_t = \text{N}(x | \mu_{S_t}, \sigma_{S_t}^2).$$
Taking the initial state $S_0$ as fixed, the unknown parameters in the model are $\theta$, $\phi$, $\mu_0$, $\sigma_0^2$, $\mu_1$, and $\sigma_1^2$.  These are the parameters that are estimated in your call of the depmix model.

Fitting output: The first six lines of your output (starting with iteration 0 ... and ending with ... logLik: -88.73058) is the model-fitting part.  In each line you can see that there is an update with an additional five iterations of the fitting process, and the log-likelihood value increases each time.  In this case the iterative algorithm converged after 25 iterations and the maximised log-likelihood is the value given at the end of this output.
Summary output: The remaining output is the summary of the fitted model.  The intercept here is the estimate of the mean of the distributions of the two states, and the sd is the estimate of the standard deviation of the distributions of the two states.  So in your fitted model, your estimated parameters are:
$$\begin{matrix}
\hat{\theta} = 0.884 & & & \hat{\phi} = 0.916, \\[6pt]
\hat{\mu}_0 = 5.510 & & & \hat{\sigma}_0 = 0.192, \\[6pt]
\hat{\mu}_1 = 6.385 & & & \hat{\sigma}_1 = 0.244. \\[6pt]
\end{matrix}$$
A: Its normal probability distribution function with mean as intercept and standard deviation as sd.
When response variable is continuous then it gives PDF else fixed matrix.
