6 - Noise Reduction Flashcards
What is noise?
Noise can be described undesirable or unwanted sound, or a stimulus that is deemed to be detrimental to the intelligibility of speech
What 4 features make speech speech-like?
Hearing aids can understand speech characteristics very well
1) Patterning
2) Modulation and rate
3) Envelope characteristics
4) Amplitude
What 2 ways helped with noise reduction before today’s digital technology?
1) Fixed directional microphones
2) Low frequency gain reduction strategies
What are 3 low frequency gain reduction strategies that were used to reduce noise before today’s technology?
1) Noise suppression switch (operated by user)
2) Multi-memory devices with low-frequency gain cut in one program
3) Automatic low-frequency gain cut with increases in input level
How do low frequency gain reduction strategies work? What is the problem with this?
- Assumes noise has mostly low-frequency energy (true for some noise, e.g., traffic, machinery)
- Main problem? Gain reduction also reduced low-frequency speech cues
What 4 ways do HAs analyze incoming signals to determine whether noise is present?
1) Analyzing the intensity level in a given channel
2) Analyzing the spectrum, modulation, and rate
3) Analyzing the redundancy of the input
4) Classification of noise aims to improve speech primarily
What things can HAs classify?
Is the input quiet? speech in quiet? Speech in noise? Speech in loud noise? Music? Noise only? Is the noise steady-state? Is the noise transient? Is the noise wind?
If speech and noise are both present, what happens?
- HAs will use the information of the SNR to determine if they must turn on noise reduction or not
- You will typically see these classifiers in premium technology
What is the HA input analysis?
- These is telling us which classifiers patients are in
- This can be helpful to tell us which environments a patient is in
How well does a HA do at separating speech from noise?
If you put the speech in noise, it can be difficult for the HA to separate the two
What are the 5 different noise reduction approaches?
1) Modulation approach
2) Spectral approach
3) Digital noise reduction
4) Gain-reduction digital noise reduction (G-DNR)
5) Traditional-digital noise reduction (T-DNR)
How does the modulation approach work?
A modulation detector is used to determine if the incoming signal is speech or noise
What is the modulation detector?
- Looks at temporal and amplitude patterns (modulations) of incoming signal and determines what it is analyzing.
- If “speech” is around 10-15 S/N or better, no gain reduction (HA interprets main signal as speech)
- Speech has greater amplitude modulation depth
- Works best if noise is steady-state and narrowband. Noise tends to have less amplitude fluctuations, and can be easily identifiable
What is narrowband noise?
When the noise is in a specific frequency band and is staying steady
How does the spectral approach work?
1) The HA tries to determine the speech spectrum, and then estimates the noise spectrum from the total signal spectrum.
2) Spectral subtraction - estimated spectrum of the noise is subtracted from the total signal
3) Speech and noise spectra must be uncorrelated
4) Works best for steady-state noise
What is digital noise reduction?
1) Hearing aids continuously analyses the incoming signal to determine if it is speech or noise, and this analysis is taking place every few milliseconds
2) If noise is detected in one or more channels, the gain can be reduced in the channel(s) where noise is present
What are the 3 challenges to digital noise reduction?
1) Speech cues in that channel are also reduced
2) Noise is often unwanted/undesired speech, but may be an important environmental cue
3) Noise is often perceptual to the listener
When is G-DNR used?
Used when noise is the DOMINANT SIGNAL
How does G-DNR work?
- Reduce noiseby reducing the gain in the channel where noise is present
- This takes TIME… Time to determine that noise is dominant, time to start reducing noise, time to settle into the maximum amount of reduction
When is T-DNR used?
Used when noise is not the dominant sign, but speech is present as well
How does T-DNR work?
- Reduce noise by reducing the gain in the channel where noise is present and separate this out from speech
- How? Noise reduction occurring between segments of speech (in between words)
- Often used in combination with G-DNR
- This takes TIME… Time to determine that noise is dominant, time to start reducing noise, time to settle into the maximum amount of reduction
Can HAs classify noise accurately all the time?
- Challenges exist
- Very difficult, especially when different acoustic signals are all occurring at the same time and may be further complicated by reverberation or room acoustics
- Hearing aids use probabilities
HA might identify the main signal as speech in noise based on the following probabilities:
- 55% chance that the signal is speech in noise
- 30% chance that the signal is music
- 15% chance that the signal is noise only
How helpful is noise reduction for patients?
- Noise reduction does not improve speech intelligibility on it’s own
- Adults prefer HA with noise reduction (they want to hear everything around them)
- More comfortable and relaxing
- Decrease in listening effort
What is the acceptable noise level and noise reduction?
- ANL = annoyance by noise
- High number = very annoyed by noise
- Low number = less annoyed by noise
- When adding a little bit of noise reduction, they perform better for an ANL test
- Significantly lower ANL scores with any noise reduction strategies than no NR.
- People with poor ANL at baseline showed greater benefits of noise reduction.