0

So I am using the Watson ExampleStreaming Speech to Text demo as a starting point from the Unity SDK here

https://github.com/watson-developer-cloud/unity-sdk

It works but the issue is accuracy.

I've read this post here How can I improve Watson Speech to Text accuracy? but have one question

By simply using my own voice to create a custom acoustic model can I expect a increase in the accuracy level? For my purposes I will ONLY be using the application with my voice.

https://cloud.ibm.com/docs/services/speech-to-text?topic=speech-to-text-customization

Bachalo
  • 6,965
  • 27
  • 95
  • 189

1 Answers1

0

Using your voice could help but you need to record a lot of sample utterances and will likely pay more for custom model. Before going that route, I recommend analyzing the mistaken words to see if there's a pattern and adding manual correction for those words in your code.

For example, if the model is confusing "two" and "too", you can add a manual replace function for this.

Adnan S
  • 1,852
  • 1
  • 14
  • 19
  • Thanks already tried your suggestion. Using a custom model with grammar file is the best solution in my case. – Bachalo May 15 '19 at 20:32