📦 Orpheus-TTS / OuteTTS or other Dataset Preparation

This repository provides tools, including a console app and Python scripts, to create datasets for Orpheus TTS. Follow the steps below to prepare your dataset efficiently. You can use this tool to create or read parquet files for other datasets.

---

Prerequisites

• Metadata File:
  Prepare a meta.csv file with the following format:

  1 | text of audio 1
  2 | text of audio 2
  ...
  

• Audio Files:
  Organize your audio files in the audio folder with filenames matching the metadata:

  1.wav
  2.wav
  ...
  

---

Steps to Create the Parquet File

1. Adjust Paths in Scripts:
   Modify the following paths in the chosen script:

   metadata_path = \"G:..*meta*.csv\"
   audio_dir = \"G:..*audio*\"
   output_parquet = \"orpheus_tr_10k.parquet\"
   

2. Run One of the Below Scripts:
   Use any of the scripts based on your preference:

   • parquet_createData2.py
   • parquet_createDataNoShard.py

   Alternatively, you can use the .NET Console App to create the Parquet file.

---

Creating the Dataset

1. Run the Dataset Creation Script:
   Use the createdataset.py script.
   Update the script to reference the Parquet file:

   ds = load_dataset(\"parquet\", data_files=\"G:...*orpheus_tr_10k*.parquet\", split=\"train\")
   

2. Save the Dataset:
   Adjust the save path in the script:

   ds.save_to_disk(\"G:....*testKarayData*\")
   

---

Outputs

• Parquet File: orpheus_tr_10k.parquet
• Saved Dataset Directory: testKarayData

---

Feel free to contribute or report issues for improvements to the dataset creation process! 🚀