Lecture 7.5: Hynek Hermansky - Auditory Perception in Speech Technology, Part 2

TL;DR

The auditory system processes speech by analyzing temporal trajectories of spectral energies, and noise reduction techniques involve using long temporal contexts, sub-band processing, and multi-stream adaptation to unknown noise.

Transcript

The following content is provided under a Creative Commons license. Your support will help MIT OpenCourseWare continue to offer high quality educational resources for free. To make a donation or view additional materials from hundreds of MIT courses, visit MIT OpenCourseWare at ocw.mit.edu. HYNEK HERMANSKY: So we have this wanted information and un... Read More

Key Insights

• 🤕 The auditory system processes speech by analyzing temporal trajectories of spectral energies, inspired by equal loudness curves and critical bands.
• 😘 Linear distortions in speech can be mitigated through techniques like reduction of low frequencies and integration within critical bands.

Questions & Answers

Q: How does the auditory system process speech?

The auditory system analyzes temporal trajectories of spectral energies to extract speech information, using techniques like equal loudness curves and critical bands.

Q: How does the system handle linear distortions in speech?

Linear distortions, such as differences in vocal tracts, can be addressed by reducing low frequencies and integrating within critical bands.

Q: How can noise be mitigated in speech processing?

Noise can be reduced by using sub-band processing, multi-stream adaptation, and selecting the best stream based on performance.

Q: What are some key techniques for noise reduction in speech processing?

Techniques include sub-band processing, multi-stream adaptation, and performance-based stream selection.

Summary & Key Takeaways

• The auditory system analyzes temporal trajectories of spectral energies to extract speech information, using techniques like equal loudness curves and critical bands.

• Linear distortions in speech, such as differences in vocal tracts, can be mitigated by using techniques like reduction of low frequencies and integration within critical bands.

• Dealing with noise in speech processing involves using techniques like sub-band processing, multi-stream adaptation, and selecting the best stream based on performance.