Projects
AudioPipeline
AudioPipeline preview screenshot - Aayush Dubey

AudioPipeline

Hindi and English text-to-speech and voice-cloning studio built on a 6 GB laptop GPU, with a showcase site of before/after audio samples and measured latency and VRAM results.

  • Hindi & English Voice Generation: FastAPI backend running Chatterbox Multilingual TTS on an NVIDIA RTX 3050 laptop GPU, with a React + Vite studio frontend

  • Tuned by Measurement: A resident model singleton cut model load from about 60 s to 41.5 s and warm previews from about 126 s to 7.3 s, with conditioning-cache hits in about 31 ms and roughly 3.5 GB peak VRAM

  • Showcase Site: Astro Starlight documentation site with before/after audio comparisons, the engineering story and a run-it-yourself guide

  • FastAPI
  • Python
  • PyTorch
  • CUDA
  • Chatterbox TTS
  • React
  • TypeScript
  • Vite
  • Astro