shariqm

if you can't view the video in the link from your location, it's on youtube too: https://www.youtube.com/watch?v=sHKopqVI_cw

We're building hearing aids that work in noisy places (AudioFocus[1], YC S19). We use novel machine learning and microphone array design to help patients hear better in loud restaurants, weddings, & family gatherings better than any other AI hearing aid.

It's a big deal because untreated hearing loss is associated with social isolation & depression and while 37M people have hearing loss in the United States, only 8M use hearing aids. Hearing in noisy places is the biggest reason for lack of adoption.

We just got our behind-the-ear (BTE) hardware prototype running and already have several excited patients. Listen to an audio recording from it here[2]. We're currently working on a pilot study with a professor in San Francisco.

If you, or someone you know, is interested in participating in the pilot study let me know. And if you know interested investors, I'm happy to chat with them. I can be reached at shariq@audiofocus.io

[1] www.audiofocus.io

[2] https://www.youtube.com/watch?v=orU5Wx6_RfA&t=24s

Those are pretty cool screen reader add-ons. Do you use JAWS, NVDA, or something else?

AudioFocus | Oakland, CA | Hardware Design Co-Founder | Full-time

AudioFocus (YC S19) makes hearing aids that work in noisy places like bars. We use a combination of an in-house built auralization framework and deep learning models to accomplish this. Our prototype runs on the iPhone and we're looking for someone to built a hardware prototype and take it all the way to mass manufactured scale.

You have experience in:

- Low Power Architecture Design

- Wireless Audio Transmission (NFMi, 900Mhz, 2.4Ghz)

- Validation & Mass Manufacturing

and are a lifelong learner, have interest in ML processors, and have worked at Apple AirPods, a hearing aid company, Bose, Doppler Labs, or something similar.

Founded by me (UC Berkeley, BS-EECS, 2011; UC Berkeley, PhD-ML, 2019). Contact: shariq@audiofocus.io

Thought this paragraph was particularly interesting:

“The notion that any risk assessment instrument can account for bias ignores the racial disparities in current and past policing practices.” There are abundant theoretical and empirical reasons to support this claim, since risk assessments are typically based on data of arrests, convictions, or incarcerations, all of which are poor proxies for individual behaviors or predispositions. The coalition continued, “Ultimately, risk-assessment tools create a feedback-loop of racial profiling, pre-trial detention and conviction. A person’s freedom should not be reduced to an algorithm.” By contrast, the Partnership’s statement focused on “minimum requirements for responsible deployment,” spanning such topics as “validity and data sampling bias, bias in statistical predictions; choice of the appropriate targets for prediction; human-computer interaction questions; user training; policy and governance; transparency and review; reproducibility, process, and recordkeeping; and post-deployment evaluation.”

Seem's like they're addressing superficially irresponsible AI deployment but ignoring some of the deeper, complex, ethical issues raised by technologies like facial recognition.

I agree it's still not going to win the cost-benefit tradeoff in most markets. OTOH - I imagine making your software support more OSs is much more expensive than designing your website to be more accessible, at most $1M/year. If you've got billions in annual revenue that's pennies in the bucket.

In my view, it's about trying to provide more 'equal opportunity' in this world where it's financially feasible. This seems like a case where it is more often that not. Again, for medium+ businesses.

Ah yeah it does say that here: https://www.eeoc.gov/laws/guidance/ada-primer-small-business

Good to know

I see some people are feeling conflicted trying to balance 1) Making the web more accessible with 2) Increasing business/startup costs.

It's important to remember making websites more accessible often benefits everyone in some way. e.g. providing an alt text description for images on your website obviously benefits visually impaired users, but it also benefits people who can't afford high bandwidth or people who are moving through a remote area. In the Microsoft framework these three scenarios are called:

1) Permanent (Visually Impaired)

2) Temporary (Can't afford high bandwidth)

3) Situational (In a remote area)

There's good online resources for understanding these concepts [1,2,3].

Anyway, for companies as big as WinnDixie, with $10B in annual revenue, they should be designing their websites to be accessible by everyone. For small businesses it would be nice if website building providers like SquareSpace and Wix made this as easy as possible for their users.

[1] https://www.iweb.co.uk/2016/10/inclusive-design-why-our-webs...

[2] https://www.youtube.com/watch?v=TAzkrXTGEOM&t=55s

[3] https://www.w3.org/standards/webdesign/accessibility

Yeah, this isn't the first time: https://dralegal.org/case/national-federation-of-the-blind-o...

Uber settled a case similar to this in 2016 [1,2] and agreed to "Require that drivers provide equal service to people with disabilities who travel with service animals." Seems like they haven't learned their lesson, and this minor new $1.1M fee is unlikely to change their behaviour. OTOH, from a Teacher of the Visually Impaired (TVI) I know: -

"It’s challenging. I think they need to make it really clear this is part of their job. It’s common for people not to allow dogs in their car, just in general. On top of that, they employ a lot of people who don’t speak english and people from several cultures who would never have a dog in the car."

[1] https://dralegal.org/case/national-federation-of-the-blind-o...

[2] https://www.ada.gov/briefs/uber_soi.pdf

Very cool! I like the idea of a longer period (1-year) to develop your idea, good science takes time.

A big hurdle for doing this is latency, every time you hop from one device to another you're adding at least ~10ms of latency if it's over anything like bluetooth. There might a clever way to do it, it's something I'm thinking about but it's too early to say.

For now it looks like it's going to be a behind-the-ear hearing aid. Maybe in 5-10 years it could fit in your ear canal.

In this demo there is only one microphone on the phone. But we're working on stereo (2) microphones now. That will help make it more reliable, lower latency, and give you spatial awareness (awareness of the direction of sound).

Glad you're excited about it!

I agree! We're working on this at AudioFocus (YC S19). You can see an old demo here: https://www.audiofocus.io/demo

There are a lot of challenges with putting this in a hearing aid: low latency inference (10ms), small power budget, consistent performance, fault tolerance, etc. Still a fair amount of R&D to be done, but change is coming.

I would definitely call that low power. Our entire brains only use 20 watts of power, compare that to the 85,000 Watts IBM's Watson uses.

As someone with a PhD in neuroscience I like the sentiment of this post but it’s off the mark in some key ways.

From a high level @foone is right, our eyes and visual system are nothing like a camera. Our visual system constructs an internal 3D model of the world, our “perception”, given observations in the form of the electrical activity in our photoreceptor (rod and cone) cells. It tries to do this in a way that balances energy use and accuracy of the model. This simple concept can actually be used to explain a lot things that may seem like ‘hacks’ as @foone put it. One example: Saying Saccades are a problem is misleading, they likely evolved to give us higher acuity (resolution) vision without having to pay the energy cost of a larger fovea [1, 2, 3]. They also likely allow us to see color more accurately outside of our fovea.

I also really don’t buy this time-shifting idea, although other neuroscientists would probably disagree. As @foone said our eyes are not cameras, our perception is our brain hallucinating an internal 3D model of the world. It knows what data to incorporate into the model and what not to, i.e. a blurry saccade is not a good representation of the world. If you want to read about how this can be framed with bayesian statistics you can read one of my favorite papers from my advisor [4].

[1] https://en.wikipedia.org/wiki/Saccade#Function

[2] Ratnam, Kavitha, et al. "Benefits of retinal image motion at the limits of spatial vision." Journal of vision 17.1 (2017): 30-30.

[3] Anderson, Alexander G., et al. "A neural model of high-acuity vision in the presence of fixational eye movements." 2016 50th Asilomar Conference on Signals, Systems and Computers. IEEE, 2016.

[4] Olshausen, Bruno A. "27. Perception as an Inference Problem." The cognitive neurosciences (2014): 295-304. https://pdfs.semanticscholar.org/bb00/42b5e48feff89a95182c63...

As someone who was often distracted from grad school research to work on grant applications this is great news. I would caution against making the NSF more like DARPA though. DARPA definitely breathes down your neck asking for monthly updates that hinders research/ideas that take longer to develop.

Federal research funding should be directed towards developing big, long-term ideas rather than short-term research tech companies can and do focus on.

Yeah, the OTC (over-the-counter) hearing aid act passed in 2017 [1]. It says the FDA has to release reduced regulations for the sale of hearing aids OTC by August 2020. There will be a bunch of new players, e.g. Bose HearPhones and maybe even Apple (checkout Live Listen).

The challenge will be how to fit a hearing aid without an audiologists help. This will be especially difficult for older people.

Hopefully it will bring down the price of hearing aids substantially though.

[1] https://www.warren.senate.gov/newsroom/press-releases/bipart...

The most common type of hearing loss is age related hearing loss. This type of hearing loss makes it difficult to hear high frequencies. The main thing a hearing aid does is compensate for this by adjusting the volume of high frequencies. But here's what you're missing - The key is to amplify at low volumes and reduce at high volumes, AKA dynamic range compression [2]. This allows people to hear better at low volumes but not blast their ears at high volumes.

FYI - The ear piercing shriek you heard was from poor feedback cancellation on the hearing aid. Hearing aids have a microphone and speaker so feedback is common and a nightmare.

Also, I'm the founder of AudioFocus (YC S19)

[1] http://www.hearingresearch.org/ross/hearing_loss/the_audiogr... [2] http://openaudio.blogspot.com/2017/01/the-need-for-dynamic-r...

There's plenty of hearing aid apps on iOS and Android. They haven't really taken off because of the delay introduced by the mobile OS and the wireless transmission (> 10ms) causing you to hear an annoying echo effect.

Sorry, I'm not following. If you have 2 rows storing how much money you have in 1 account then your data isn't normalized and serializability isn't going to solve your problem, if you could be (even) more explicit, I might be able to answer.

"MVCC can't stop two competing transactions debiting the same account pair (i.e., I have two linked accounts A & B with $100 in each and I launch two simultaneous transactions to transfer $150 away from both: afterwards, I'll end up with -$50 in both or some garbage data)."

That isn't true. In a pure MVCC world you could detect such issues, specifically that two trxs were opened and they both tried to update the same row. This would generate a conflict and one transaction could just get shot. Snapshot-isolation/MVCC describes this issue.

However what InnoDB/mysql and Clustrix actually do is use MVCC for read-write conflicts and two-phase-locking for write-write conflicts which trivially fixes your proposed problem.

Clustrix says it provides repeatable-read because it complies with the ANSI spec. In truth we actually provide something closer to snapshot-isolation by default. Snapshot isolation in summary is a guarantee that all reads made in a transaction will see a consistent state of the database. A transaction will only fail to commit if there is a conflict of two updates. This is a strong isolation level that does have a set of anomalies but most companies don't care about them. Microsoft put out a great paper about snapshot-isolation: http://research.microsoft.com/pubs/69541/tr-95-51.pdf

As the post points out, weaker isolation levels allow for more concurrency which translates to more scalability, which is why we chose our isolation level.