# ONNX ASR & Diarization Pipeline

A high-performance audio transcription and speaker diarization tool utilizing ONNX-optimized Parakeet TDT models and Pyannote Audio.

## Features
- **Fast Transcription**: Uses ONNX-based Parakeet models (FP16/INT8/FP32).
- **Speaker Diarization**: Integrated with `pyannote.audio` (v3.0/3.1).
- **Evaluation Metrics**: Automatically calculates WER, CER, DER, and WDER if ground truth is available.
- **Reporting**: Generates detailed text reports including Real-Time Factor (RTF) metrics.

## Prerequisites
- **FFmpeg**: Must be installed on your system path to handle audio conversion.
- **Hugging Face Token**: Required for Pyannote models. Accept the user terms for [pyannote/speaker-diarization-3.1](https://huggingface.co/pyannote/speaker-diarization-3.1) on Hugging Face.

## Installation

1. **Clone the repository:**
   ```bash
   git clone <your-repo-url>
   cd <repo-name>

2. **Create a virtual env:** 
    ```bash
    conda create env -n <envname>
    conda activate <envname>

3. **Install Dependencies:**
    ```bash
    pip install -r requirements.txt

4. **Set HuggingFace Token:**
    ```bash
    export HF_TOKEN="your_token_here"