Natural language processing · Speech recognition

Are there start ups focused on speech to text or had this been left largely unexplored?

Sreyantha M Sprint Fellow at Public Laboratory for Open Technology & Science

February 27th, 2015

While I am trying to start one, I could not find any start up in this technological area. Why is it left unexplored? Or am I missing something?

EDIT: I would really appreciate if you could tell me what it is that I am missing in this context. Also, I would love to know about the start ups that are focused on Non-English scripts.

EDIT 2: This question is closely related to "Speech Recognition" than to NLP.

Sid Thekkepat VC, Startups

February 27th, 2015

Speech recognition is a really hard problem to crack. i.e. simply using open source recognizers/engines only gets you to about 70-80% accuracy. It takes a lot more engineering time and data! to then get to 90% accuracy levels to build a robust application. This makes it a space that larger players - Google, Nuance, MSFT, Apple and Facebook are better suited to crack. 

However, not to say that there aren't niches to explore. 
- There are lots of applications that can be built on the back of existing Google/Nuance engines (Nuance is pricey and not startup friendly btw!)
- You rightly point out that ASR (Automatic Speech Recognition) in non-English (for e.g. Asian languages or under resourced languages as they are referred to) has lots of potential (requires time and data though)
- If you're keen on focusing on still building out a speech recognition new engine, I'd suggest focusing on areas where 70-80% accuracy levels are enough. Happy to share more ideas if you want to know them.

Ian Saunders Co-founder of CryoAction, cryotherapy chambers for elite sports. Startup veteran & advisor

February 27th, 2015

There certainly are a great many startups in the speech to text technology space. In Europe there is an organisation called LT-Innovate, a forum to support such organisations. They produce a directory of European companies in this space- http://lt-innovate.preprod.dotprojects.be/directory

Jeff Chang Startup Guy, ER Radiologist & Hunting for AI

February 27th, 2015

For a quick background:

http://en.wikipedia.org/wiki/Speech_recognition / http://en.wikipedia.org/wiki/Natural_language_processing

http://en.wikipedia.org/wiki/List_of_speech_recognition_software

(Early) IBM in 2007:
http://www.infoworld.com/article/2648372/applications/ibm-develops-speech-recognition-in-indian-language.html

Google in 2014:
http://tech.firstpost.com/news-analysis/google-translate-gets-voice-recognition-for-hindi-and-seven-other-indian-languages-225225.html

Ben Sweat Director, Product at Idealab

February 27th, 2015

Big companies are doing this. Google and Nuance communications (Dragon Dictation) are the leaders. It's really freaking hard and they are already really good at it, which is why you don't hear about startups. IMO. Nuance handles speech to text for Apple. Open Ears is the open source alternative. It's ok. But all of the acoustic models they use are apparently just adult males because it doesn't work well for women or children. I've built an app with this. ToyTalk raised a bunch of money to create an engine to do part of this. They have many engineers and smart people in this space. What I'm not sure of is how much of the speech recognition they do themselves or if they built on top of Dragon. Their product they call a conversation engine for interactive apps and toys. Sent from my iPhone

Karl Schulmeisters CTO ClearRoadmap

February 27th, 2015

You need to add Microsoft into that as they have  commercial AVR system as well as the demo of on the fly voice to voice translation in Skype that they showed https://www.youtube.com/watch?v=rek3jjbYRLo

After all if they can parse inbound voice
identify meaning
and then translate and respeak it

the output part of going to text instead of speech is trivial.

And yes this requires a lot of processing.  Basically Google, Microsoft and perhaps Amazon have the horsepower to do this.  

Don't bet against them unless you already have some brilliant research insight that is groundbreaking at the academic level

Greg Sherwin vp engineering + it • singularity university

February 28th, 2015

TBH, it sounds like there are gaping holes in your market research, IMO. It's not just the big players like Google and Nuance. There are others like Amazon's INOVA, LumenVox, CereProc, Cepstral, Lessac Technologies, iSpeech, Acapela, etc. This isn't just a market that is explored, but it is arguably saturated.

Darrell Jr. Digital Marketing Consultant, Writer and Entrepreneur

February 27th, 2015

Most of this work is being done via machine learning... You'd be competing with the likes of Google with their untold processing power. Needless to say this is out of reach for most startups.

Kerry Davis

February 27th, 2015

For me, the immediate need in this area is already provided by my cellular carrier. So I think your problem will be that even if you identify a very compelling use case, there will be little if any barrier to entry for the competition.

Jeh Daruvala Founder & CEO - Yactraq Online Inc.

February 27th, 2015

Hi Sreyantha, please take a look at www.yactraq.com. We are a speech based startup. Best to take the conversation offline if you are interested in talking further. To contact me please send a private message to me directly on FounderDating - Jeh Daruvala. If for some reason that does not work you can message me directly via bizdev at yactraq.com

Karl Schulmeisters CTO ClearRoadmap

February 28th, 2015

Good Point Greg!  there are domain specific tools a plenty.