As the digital music library continues to grow, the quantity of newly released songs gradually increases. So, there is a rising demand for an automated system capable of categorizing each song, providing significant information like the artist or genre. A proficient singer identification model can recognize a singer who is singing by examining a piece of audio clip. This paper concentrates on signer identification in classical music. On account of this, a novel metaheuristic optimization-based hybrid model for singer identification model is proposed in this research work. And this model executes effectively by following three phases—vocal segmentation, feature extraction, and singer identification. The input audio is given into Bi-LSTM model in vocal segmentation phase which splits the given audio into vocal and instrumental regions. From feature extraction phase, features such as STFT, MFCC, and chroma features are retrieved from the vocal patterns. Subsequently, the singer identification phase is executed using a hybrid model. This hybrid model is formed by the combination of fine-tuned CNN and fine-tuned Bi-GRU models where these models are fine-tuned by a novel Self-Improved Black Widow Optimization (SI-BWO) algorithm. Hence, this optimization-based hybrid model provides more accurate prediction results on singer identification. Further, the effectiveness of proposed singer identification model is evaluated by various experimental analyses and its significance is validated over existing singer identification models.