A Corpus Linguistics-Driven Hybrid Deep Representation Framework Incorporating Majority Voting for Enhanced Automatic Spoken Language Identification

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Spoken Language Identification (SLID) is a well-researched area. It has previously been recognized as a crucial first stage in every multi-lingual speech recognition system. Latest developments in artificial intelligence (AI) and deep learning (DL) are used to highly strengthen the performance of SLID systems, allowing them to perform a crucial part in mainstream applications The gathered speaker utterances have been entered into the DL network for training phase. The language equivalent to the great posterior probability is then recognized as the target language. This study introduces an Enhancing Automatic Spoken Language Identification with Corpus Linguistics and Ensemble Deep Representation (EASLI-CLEDR) method. The motive of the developed EASLI-CLEDR approach is to classify and identify the incidents of spoken language. Primarily, the proposed EASLI-CLEDR approach performs comprehensive audio preprocessing, including resampling, framing, and windowing, normalization, and noise reduction to effectively remove background noise and enhance signal quality. Next, spectrograms can be produced from the speech data rather than directly treating the voice data to mine additional discriminative aspects. Additionally, a majority voting-based hybrid DL approaches integrating gated recurrent unit, temporal convolutional network, and graph convolutional network models are utilized for carrying out the SLID classification process. Finally, the parameters of the EASLI-CLEDR technique are effectively optimized using the Adadelta optimizer, resulting in enhanced recognition performance. To confirm the enhanced performance of the EASLI-CLEDR technique, experiments were conducted using the Identification of Language in an Audio dataset. The comparison study reported that the EASLI-CLEDR approach outperformed existing techniques using multiple evaluation metrics.