International business machines corporation (20240161733). GENERATION OF TRAINING EXAMPLES FOR TRAINING AUTOMATIC SPEECH RECOGNIZERS simplified abstract

From WikiPatents
Jump to navigation Jump to search

GENERATION OF TRAINING EXAMPLES FOR TRAINING AUTOMATIC SPEECH RECOGNIZERS

Organization Name

international business machines corporation

Inventor(s)

Ngoc Minh Tran of Dublin (IE)

Hessel Tuinhof of Dublin (IE)

Beat Buesser of Dublin (IE)

GENERATION OF TRAINING EXAMPLES FOR TRAINING AUTOMATIC SPEECH RECOGNIZERS - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240161733 titled 'GENERATION OF TRAINING EXAMPLES FOR TRAINING AUTOMATIC SPEECH RECOGNIZERS

Simplified Explanation

The present invention involves a method, computer program product, and computer system for generating training examples for training an automatic speech recognizer. The method includes receiving a training dataset of original audio signals and generating training examples based on a constructed imperceptible space for the original audio signals and adversarial audio examples in the constructed imperceptible space. An imperceptible and adversarial audio example is then generated for an adversarial trainer for the automatic speech recognizer.

  • Training examples generated for training an automatic speech recognizer
  • Based on imperceptible space for original audio signals and adversarial audio examples
  • Imperceptible and adversarial audio examples created for training
  • Utilizes a training dataset of original audio signals

Potential Applications

The technology can be applied in:

  • Improving the accuracy and performance of automatic speech recognition systems
  • Enhancing the robustness of speech recognition in noisy environments

Problems Solved

This technology addresses:

  • Challenges in training automatic speech recognizers with diverse audio signals
  • Overcoming limitations in traditional speech recognition training methods

Benefits

The benefits of this technology include:

  • Increased accuracy and reliability of automatic speech recognition systems
  • Improved adaptability to different audio environments
  • Enhanced overall performance of speech recognition systems

Potential Commercial Applications

With its capabilities, this technology can be utilized in various commercial applications such as:

  • Voice-controlled devices and virtual assistants
  • Call center automation systems
  • Speech-to-text transcription services

Possible Prior Art

One possible prior art in this field is the use of adversarial training techniques in machine learning to improve model robustness and performance.

Unanswered Questions

How does this technology compare to traditional speech recognition training methods?

This technology improves upon traditional methods by incorporating imperceptible spaces and adversarial examples for more effective training. It would be interesting to see a comparative study showcasing the performance differences between the two approaches.

What impact does the imperceptible space have on the generated training examples?

The concept of imperceptible space is crucial in this technology, but the specific effects and benefits it provides to the training examples could be further explored and explained in detail.


Original Abstract Submitted

a method, computer program product, and computer system for generation of training examples for training an automatic speech recognizer. embodiments of the present invention can receive a training dataset of original audio signals and generate training examples for training an automatic speech recognizer based, at least in part, on a constructed imperceptible space for an original audio signal of the original audio signals and adversarial audio examples in the constructed imperceptible space. embodiments of the present invention can then generate an imperceptible and adversarial audio example to an adversarial trainer for the automatic speech recognizer.