Answers to some of your queries are as follows:

- It depends on network configurations, the way one creates training examples & datasets. A lack of systematic benchmarking of existing methods however creates confusion. There are several studies which shows that LSTM outperforms CNN for speech emotion recognition. Specially LSTM with attention mechanism helps to boost emotion recognition performance. Some other studies report the opposite and CNN seems to be a better choice. You can do a quick literature survey on latest papers published in last couple of INTERSPEECH, ICASSP & ASRU.

- MFCC is the default choice for most speech processing tasks including speech emotion recognition. However, MFCC is not the optimal one as it lacks prosody information, long-term information. That's why often MFCC is augmented with pitch (to be more specific log F0) and/or shifted delta coefficients. These additional information help to boost the emotion recognition performance. MFCC lacks phase information but the role of phase in emotion recognition performance is not much investigated. The parameters for MFCC computation such as the number of filters, the frequency scale are chosen experimentally and they are dependent on the dataset and the backend classifier.

-The answer is yes. I believe it is practically and theoretically proved that CNN is a more suitable network for applications that use time-frequency domain signals and spectrograms. If the signals are analyzed in the time domain or frequency domain RNN would be more handy.

-RNN is more efficient on vector inputs and not matrices. I can't really see any reason to choose RNN over CNN when the input is a matrix.

-Like any other deep network optimizing the parameters and the network configuration (convolution layer) is the most challenging task when using CNNs.

The RNN is more effective for SER while CNN for image and video

The MFCC is used for SER because it relates to the human