Project Pingala: Speech Synthesis for Sanskrit Verse

Description

Sanskrit is a versatile language consisting of various meters and tunes. It’s recent involvement with artificial intelligence (and recognition from NASA) is an honor proving its power for being a valuable course of literature. The synthesizer takes in Sanskrit text as an input, considers the “meter” in which a given verse is composed, and modulates the syllables accordingly to generate speech. Many great Sanskrit scholars struggle to recite verses written using meters like Āryā, which are tough to recite. There has been close to no work on Text-to-Speech models in Sanskrit. This tool would aid scholars and learners in improving their pronunciation of Sanskrit verses. We are also planning to create a dataset publicly to encourage more research in this field. Future goals include generalizing the approach adopted here to other languages like Tamil, Telugu, etc. The below example depicts how very similar-looking verses are pronounced entirely differently meters.

अस्मिन्विकारःखहरे न राशावपि प्रविष्टेष्वपि निस्सृतेषु।
बहुष्वपि स्याल्लयसृष्टिकाले अनन्तेऽच्युते भूतगणेषु यद्वत्।।

सैकपदघ्नपदार्धमथैकाद्यङ्कयुतिःकिल सङ्कलिताख्या।
सा द्वियुतेन पदेन विनिघ्नी स्यात्तृहृता खलु सङ्कलितैक्यम्।।

भाज्यो हारः क्षेपकश्चापवर्त्यःकेनाप्यादौ सम्भवे कुट्टकार्थम्।
येनच्छिन्नौ भाज्यहारौ न तेन क्षेपश्चेत्तद्दुष्टमुद्दिष्टमेव।।

Application Process

To apply for this project one must attempt to solve a (simple) selection task and state why you want to be part of the project. The task involves parts:

Interest Form

Asp. PM WhatsApp Group Link

Application Link

TIP: Deep Learning was invented to mimic how a neuron/human learns (atleast to a certain degree). Keep this in mind while answering the questions above. Your answers need not be technically accurate (but it would be appreciated) and can be loosely worded, but they MUST have some intuition backing them.

Submission Deadline

Completed applications are to be submitted by 14th May 2022 11:59PM!

Submission Link