sushant-t/tts-trainer

Generate audio datasets for training Text-To-Speech models, through smart audio splitting with silence detection, and transcription using Whisper.

TTS Trainer

Train Text-To-Speech models using automated dataset generation techniques, such as smart audio splitting with silence detection, and transcription using Whisper.

The datasets generated are in the LJSpeech single-speaker dataset format: https://keithito.com/LJ-Speech-Dataset

Requirements

Mamba Package Manager: https://github.com/conda-forge/miniforge#mambaforge

Before running, make sure you go through all of the README files located in this repository. You can find them linked below:

1. Diarization Model
2. Transcription Model
3. Input Samples

Instructions

Generating Data

python -m src.bin.generate_data --input_folder <INPUT FOLDER NAME>

Example: for speaker LJ001, run python -m src.bin.generate_data --input_folder LJ001

Training

Information coming soon.