Vāgdhenu — A Vṛtta-Aware Śloka-to-Chant TTS for Sanskrit
Try it
Paste a Sanskrit verse in any Indian script — the meter is detected automatically.
First chant takes ~10–60s while the model warms up. If the demo doesn't load,<br>use the backup demo ↗.
Listen — sample chants
Six vṛttas rendered by this system — including verses from the shipped deployments.
vasantatilakā<br>Mahābhārata Tātparya Nirṇaya · 1.1
नारायणाय परिपूर्णगुणार्णवाय विश्वोदयस्थितिलयोन्नियतिप्रदाय ।<br>ज्ञानप्रदाय विबुधासुरसौख्यदुःखसत्कारणाय वितताय नमोनमस्ते
śārdūlavikrīḍita<br>Śrīmad Bhāgavatam · 1.1.2
जन्माद्यस्य यतोऽन्वयादितरतश्चार्थेष्वभिज्ञः स्वराट् तेने ब्रह्म हृदा य आदिकवये मुह्यन्ति यं सूरयः। तेजोवारिमृदां यथा विनिमयो यत्र त्रिसर्गो मृषा धाम्ना स्वेन सदा निरस्तकुहकं सत्यं परं धीमहि
anuṣṭubh<br>Śrīmad Bhāgavatam · 1.1.5
नैमिषेऽनिमिषक्षेत्रे ऋषयः शौनकादयः। सत्रं स्वर्गाय लोकाय सहस्रसममासत
vaṃśastha<br>Śrīmad Bhāgavatam · 1.3.5
पश्यन्त्यदो रूपमदभ्रचक्षुषः सहस्रपादोरुभुजाननाद्भुतम्। सहस्रमूर्धश्रवणाक्षिनासिकं सहस्रमौल्यम्बरकुण्डलोल्लसत्
drutavilambita<br>Śrīmad Bhāgavatam · 1.1.4
निगमकल्पतरोर्गलितं फलं शुकमुखादमृतद्रवसंयुतम्। पिबत भागवतं रसमालयं मुहुरहो रसिका भुवि भावुकाः
mālinī<br>Narasiṃha stuti · retroflex tongue-twister
हठलुठ दल घिष्टोत्कण्ठदष्टोष्ठ विद्युत् सटशठ कठिनोरः पीठभित्सुष्ठुनिष्ठाम् ।<br>पठतिनुतव कण्ठाधिष्ठ घोरान्त्रमाला दह दह नरसिंहासह्यवीर्याहितं मे ॥
About
Vāgdhenu maps a metrical verse to its chanted pārāyaṇa recitation. Its voice is a<br>flow-matching TTS backbone retrained on a purpose-recorded, carefully designed<br>single-speaker Sanskrit chant corpus (~5 hours), with a further voice-steering retrain;<br>the neural vocoder is likewise fine-tuned for the chant register. Around the trained model sits<br>the machinery a faithful Sanskrit chant pipeline needs: a script-aware frontend that routes<br>Sanskrit through Kannada orthography (avoiding the Hindi schwa-deletion that Devanagari triggers);<br>visarga sandhi with the jihvāmūlīya and upadhmānīya allophones; the aspiration contrast; the three<br>sibilants and the full retroflex series kept distinct; homorganic anusvāra and vocalic ṝ; and a<br>vṛtta-aware mechanism that detects the meter and selects a matched reference under the<br>half-reference rule. The retrained model reaches an expert MOS of about 4.6 ,<br>and dense conjuncts — including retroflex aspirates — render correctly, the class earlier<br>architectures could not crack.
Deployments
This system produced two corpora at scale.
● Mahābhārata Tātparya Nirṇaya — 32 chapters, 5,183 verses (~17.5h) ·<br>video series ↗
● Śrīmad Bhāgavatam — ~18,000 verses across 12 books ·<br>karaoke-video series ↗