README.md

metadata

license: apache-2.0
task_categories:
  - text-to-speech
language:
  - hi
  - en
tags:
  - hindi
  - hinglish
  - tts
  - corpus
  - kupe
pretty_name: kupe-tts
size_categories:
  - 10K<n<100K

kupe-tts

Kupe TTS workspace: DhVaani-0.5 local CPU benchmark + Hindi/Hinglish TTS text corpus generation.

Contents

• benchmark.py — DhVaani-0.5 CPU latency benchmark
• reference.wav / outputs/ — clone reference + sample outputs
• tts_data_gen/ — Sarvam (gemma4) corpus generator, costs CSV, EDA
  • hindi_tts_corpus.jsonl — ~25k utterance corpus
  • generation_costs.csv — per-batch token/cost log
  • cost_eda_report.html — cost EDA dashboard

Secrets

Set env vars locally (do not commit keys):

export SARVAM_API_KEY=...
export HF_TOKEN=...