TTS - Language specific evals and models suggestions (1-2 pager)

[PritamSGB]

@rajagopalmotivate

• As per discussion, wrt further planning for TTS model’s finalization across required languages,
  - Languages to cover - Hindi, English, Telugu, Assamese, Oriya, Gujarathi
  - For each of these languages, lets have a table (same like STT 1 page by Prajna) with 3-5 models with metrics mentioned above, basis this we will finalize and integrate any required models (Gemini will soon be live for TTS anyhow), And same can serve as benchmarking for NGO use cases
  - We will capture (semantic similarity via llm), WER, latency as automated metrics, (capture cost if possible)
  - Let’s try to cover at least 100 samples

reference: https://discord.com/channels/1014768296257654865/1463434906712670446/1471472813411012723