Evaluation: STT evals (POC done)

[kartpop]

Is your feature request related to a problem?
We currently don't have STT (Speech-to-Text) evaluation capabilities in our platform. NGOs need a way to evaluate transcription
quality by comparing model outputs against ground truth.

Describe the solution you'd like
Integrate STT evaluation into the platform. We've already experimented with Gemini and OpenAI for transcription and have promising
results.

Additional Context

• New results sheet
• Initial results sheet
• Bhashini vs OpenAI vs ElevenLabs research

Next Steps

• Productionize the experimental code
• Enable NGOs to run STT evaluations on their audio datasets