AI Briefing
KOSign in

Ninada v5.1.0 Open-Sources 48 In-Browser TTS Specialists for Sanskrit and Indic Languages

·2026.11.06 00:38

Key point

The project uses a 21.7 MB iSTFTNet vocoder for mobile devices and a 27.9 MB HiFi-GAN for desktops to achieve 0.75x real-time factor synthesis.

Details

  • Ninada v5.1.0 is an open-source (MIT code, CC-BY-4.0 models) in-browser TTS system supporting Classical Sanskrit, English, and 10 Indic languages with 48 native-script 24 kHz specialists.
  • Mobile performance is optimized via a 21.7 MB distilled iSTFTNet vocoder that achieves 538 to 649 ms synthesis time (0.75x RTF) on phone WASM, avoiding the 6-14 second cold start and 1.48x RTF issues seen with WebGPU.
  • A binary ONNX protobuf splitter reduces browser RAM usage by 72% (from 584 MB to 162 MB) by slicing VITS buffers into prior, flow, and windowed vocoder components.
  • Audio quality is reported at 3.09 to 3.22 dB studio MCD across 480 held-out utterances, with a 0.00 dB quantization penalty using mixed W8A32 encoder/flow and FP16 vocoder.
open-sourcetext-to-speechindic-languageswasmonnxbrowser-ml
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

HuggingFace BlogGradium Details Phonon TTS: On-Device Speech Model Tops CPU Streaming BenchmarksPyTorchKRKOMOS Achieves SRCC of 0.900 in Korean Speech MOS PredictionGeekNewsHerdr releases native desktop app for managing terminals and coding agentsHacker NewsRivet Launches Open-Source Actor Platform for Scaling AI Agents on S3GitHub BlogGitHub adds support for ACCESSIBILITY.md files in repositoriesRedditPipesHub Ranks #1 on EnterpriseRAG-Bench with 93.24 Overall Score