BANI: A Real-Time Bidirectional Voice Translation System for Indian Languages — VAD-Segmented Streaming Pipeline, 500-Phrase Cache, and Bridge Call Architecture

Read full text Summarise with VIRAT
Authors: Anil Kumar Sharma
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper constitutes a defensive publication establishing prior art for the BANI system — a real-time bidirectional voice translation platform designed for Indian language communication. BANI (बाणी — Sanskrit: voice, speech, word) was designed and built in December 2025 by PowerPbox IT Solutions Pvt. Ltd. This publication establishes a prior art timestamp for the technical innovations described herein. No patent protection is sought; the purpose is to place this work in the public domain such that no subsequent party may obtain patent rights over the described inventions. BANI provides real-time bidirectional voice translation across all 22 official Indian languages and 10 international languages — 32 languages total. The system is architected around three principal innovations: (1) a VAD-segmented streaming pipeline that performs Voice Activity Detection to segment live audio and dispatch chunks for parallel STT (Speech-to-Text), translation, and TTS (Text-to-Speech) processing; (2) a 500-phrase pre-translated cache that reduces latency from approximately 400ms (API call) to approximately 2ms (cache hit) for common phrases — a 200x speedup — using fuzzy matching for phrase variations; and (3) a bridge call architecture enabling one participant on a phone call (Twilio IVR) and one on the BANI application to conduct a fully translated conversation in real time, with WhatsApp (Baileys) integration as a third participation modality. The system employs a pluggable provider model: STT providers include Sarvam, Whisper, and IndicWhisper; translation providers include Sarvam, IndicTrans, and Google; TTS providers include Sarvam, Piper, XTTS, IndicF5, and Edge. BCP-47 language codes are mapped to Sarvam API codes. All session, latency tracking, and parallel processing logic is described in sufficient technical detail to enable reproduction.

Fields of Study & Subjects

Speech Recognition and Synthesis Natural Language Processing Techniques Speech and dialogue systems

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN
Document type
Preprint