feat: benchmarking separate models

Summary

Implemented benchmarking for all major models used in the ASR pipeline by isolating and evaluating each model independently to analyze performance characteristics.

Models Benchmarked

Transcription Models

• swecha_gonthuka
• distil-whisper/distil-large-v3

Speaker Diarization

• pyannote/speaker-diarization-3.1

Punctuation Restoration

• ModelsLab/punctuate-indic-v1

Language Recognition

• openai/whisper-small

Benchmarking Metrics

The following metrics were collected for each model:

• Model Load Time
• Transcription/Inference Time
• RAM Usage / Memory Consumption

Benchmark Dataset

Benchmarking was performed using both Telugu and English audio samples with varying durations:

• 30 seconds
• 60 seconds
• 1 minute

Purpose

The benchmarking was conducted to:

• Measure individual model performance in isolation
• Compare inference efficiency across models
• Analyze memory utilization and scalability
• Identify performance bottlenecks in the ASR pipeline

Notes

• Each model was benchmarked independently to avoid interference from other pipeline components.
• Results can be used for optimization, deployment planning, and future model selection decisions.

*closes #23