Vāgdhenu is a corpus of recorded Sanskrit chanting by a single person, primarily intended for Sanskrit speech synthesis training, prosody and meter research, and language accessibility applications. This dataset contains 1,467 segments, with a total audio duration of approximately 5.3 hours. It is in 24 kHz mono WAV format and includes two independent subsets, style_a and style_b, covering a large number of different verses. Audio processing follows the rule of no inter-sentence pauses in breathing, highlighting the pronunciation characteristics of specific Sanskrit phonemes, strictly adhering to the norms of traditional classical recitation, and excluding Vedic pitch variations.

Dataset composition:

• style_a: 764 clips, approximately 2.70 hours
• style_b: 703 clips, approximately 2.64 hours

Data fields:

• Public fields: file_name (filename), text_devanagari (Devanagari Sanskrit text), text_slp1 (SLP1 Romanized transliteration), text_kannada (Kannada text), session (recording session), take (number of recorded segments)
• The style\_a field has the following specific characteristics: duration (segment duration) and deva (source text category).
• The style\_b field contains the following specific fields: meter (rhythm and meter) and n\_syll (number of syllables).

Citation

Both the audio and recordings are the property of the author, Vāgdhenu. Please cite Vāgdhenu.