This study investigates the novel utilization of the XLSR- Wav2Vec2 model for automatic speech recognition (ASR) in Sanskrit, a language of profound historical and cultural importance. It capitalizes on the cross-lingual transfer abilities of the XLSR-53 pre-trained framework, optimizing this on the meticulously curated Sanskrit audio dataset, employing the Wav2Vec2 processor for feature extraction and training with connectionist temporal classification (CTC) loss. To determine the practical utility of our fine-tuned system, we conduct a number of experiments and measure validation loss, training loss, and Word Error Rate (WER) throughout several epochs. The results demonstrate that cross-lingual transfer learning is impactful in addressing the linguistic challenges for under-resourced languages, as seen by a significant improvement in Sanskrit ASR accuracy. This work encourages a wider use of automated speech recognition (ASR) technology in a range of linguistic settings by advancing the field of ASR for Sanskrit and providing a framework that may be adapted for additional low-resource languages.