Google STT vs nvidia Riva vs Kaldi for customising ASR

Question

I am trying to build a customized ASR there are three options I could think of to implement this.

1. Google STT
2. Nvidia Riva
3. Kaldi ASR

Which is best for customizing?

Answer 1

Nvidia Riva and Kaldi ASR are both much more customizable than Google STT. However, the latter is much easier to use, and still allows a small degree of customization such as hint phrases.

Answer 2

Google STT:

• Google STT is a cloud-based speech recognition service that offers excellent accuracy and supports a wide range of languages. It is easy to use and has a straightforward API. However, you can't customize Google STT. You can only set some arguments of API.

Nvidia Riva:

• Nvidia Riva is a GPU-based platform for building and deploying speech recognition models. It offers excellent performance and flexibility, and you have full control over the training process. However, Riva can only be deployed on supported models.

Kaldi ASR:

• Kaldi is an open-source toolkit for speech recognition that offers a wide range of customization options. It supports various acoustic and language models and provides a flexible framework for building custom models. However, most models are focused on research, not production.