Timniel/naija-asr-hf-uploader

# Naija ASR Hugging Face Uploader This repository contains a Python automation script designed to package processed audio data and transcripts into a Hugging Face `Dataset` object and push it directly to the Hub. It is specifically configured to publish the **Naija-ASR-Corpus v1.0 (NAC-v1.0)**, mapping local audio files to a transcript CSV/Excel file. ## 🚀 Features * **Google Drive Integration**: Mounts and reads data directly from cloud storage. * **Path Mapping**: Automatically matches transcript row indices to filename patterns (e.g., Row 1 -> `pidgin_0001.wav`). * **Validation**: Checks if all audio files referenced in the transcript actually exist before processing. * **Audio Casting**: Converts file paths into Hugging Face `Audio` features (resampled to 16kHz). * **Automated Documentation**: Generates and uploads a formatted `README.md` (Model Card) with YAML metadata directly to the Hub. ## 📋 Prerequisites To run this script, you need: 1. **Hugging Face Account**: You need a Write Token. 2. **Processed Data**: * A folder of `.wav` clips (e.g., `clips/`). * A transcript file (`.csv` or `.xlsx`) where the first column contains text. 3. **Environment**: Google Colab (recommended) or a local Python environment with `ffmpeg` installed. ## 🛠️ Installation & Setup ### 1. Dependencies The script installs the necessary libraries automatically, but if running locally: ```bash pip install datasets huggingface_hub librosa soundfile pandas openpyxl ``` ### 2. Configuration Open `main.py` (or your notebook) and update the **Configuration Section** at the top: ```python # Path to folder containing 'pidgin_0001.wav' AUDIO_FOLDER_PATH = "/content/drive/MyDrive/clips" # Path to your Excel or CSV file TRANSCRIPT_FILE_PATH = "/content/drive/MyDrive/transcripts.xlsx" # Your Hugging Face Repo ID (Must create this empty repo first or allow script to create it) REPO_ID = "your-username/Pidgin_ASR_Dataset" ``` ## 🏃 Usage 1. **Run the script**: ```pyt …