In recent years, speech emotion recognition (SER) and emotive talking head generation have drawn significant attention. However, existing methods either focus predominantly on emotion recognition without considering its intensity or produce emotionally consistent talking heads, neglecting the rich expressiveness of human interactions. Addressing these limitations, our study aims to bridge these two areas, proposing an innovative talking-head generation approach that further enhances the realism of the generated video. Within the SER domain, we develop a dual-level model that not only improves the accuracy of identifying emotions in speech data but also assesses their intensity. This dual-capacity model significantly enhances the depth and subtlety of emotional recognition, achieving an industry-leading accuracy rate of 73.5%. On the talking head front, this work generates emotionally expressive talking heads that reflect both emotional state and intensity. Beyond this, the model also simulates human facial dynamics like blinking, thus offering a more comprehensive and realistic representation of human speech. These advances effectively address the shortcomings of existing methods and provide a new avenue for the accurate portrayal of emotional intensity in synthesized talking heads. Extensive experimental results demonstrate the effectiveness and superiority of our approach over existing methodologies in both SER and talking head generation.