Prepare a Dataset
Recording High-Quality Datasets
30-60 minutes of clean and varied audio will result in the highest-quality voice models.
Clean audio
The highest-quality voice models are recorded:- with a quality microphone into an audio interface
- in a noise-free setting with limited room reverberations
- with correct and consistent mic placement
- with volume peaks between -9db and -3db
- with consistent dynamics across the whole dataset
- with light EQ to remove any muddiness, hiss, etc.
- with compression/limiting to smooth out peaks
- with no reverb, delay, or doubling