2025 03 25 dhwani 4 weeks march

Dhwani - 4 week - Experiments - March 2025

Below is brief summary

  1. Parler-tts: inference speed improved from 2s /word to .5 s / word.

  2. latency report : https://github.com/sachinsshetty/onwards/blob/main/idea%2Fdhwani%2Fserver%2F2025-03-16-tts-latency.md

  3. Pull Request on parler-tts github repo to enable fast inference.

  4. Current version of transformer not able to utilise speed up provided by pytorch.
  5. Updated to transformer v.50.0 and fixed deprecated functions

  6. https://github.com/huggingface/parler-tts/pull/206

  7. Conducted workshop at Chanakya University on 20th march. Topic - Getting started with Dhwani.

  8. Recording: YouTube - https://youtu.be/f5JkJLQJFGA

  9. Slides: https://tinyurl.com/dhwani-workshop

  10. Source Code: https://github.com/slabstech/dhwani-workshop

  11. Dhwani API server - GPU utilization

  12. to maximize GPU compute and use spare capacity, created API endpoints and made it available for Workshop Attendees to bootstrap projects .
  13. https://youtu.be/RLIhG1bt8gw