AdroitAnandAI/Indian-Accent-Speech-Recognition

Implements both classical signal processing pipelines (MFCC feature extraction with HMM/DTW) and modern deep learning approaches, including custom CNN-RNN architectures and fine-tuned Baidu DeepSpeech models trained on 50+ GB of multi-state Indian speech data from IIT Madras. Provides pre-trained models with KenLM language model integration and demonstrates superior performance on Indian-accented English through WER, WACC, and BLEU score comparisons against baseline DeepSpeech. Includes end-to-end training notebooks for both traditional ASR and transfer learning workflows with configurable hyperparameters.

No commits in the last 6 months.