There are a million and one services for voice transcription in the marketplace. but even with only 1 job to do, I’ve on no account viewed a service that can address the long tail of vocabulary used in the real world. here is above all difficult in case you’re a startup making an attempt to promote your provider to companies that count on correct transcription for his or her operations.
Jon Goldsmith, co-founding father of Tetra, a voice transcription startup, knows this challenge — in fact, he’s even willing to confess that he hasn’t 100% cracked the difficulty. but Goldsmith believes the reply lies in deep learning, and he’s setting out to show it with a $ 1.5 million seed round led by way of make bigger partners, with participation from Y Combinator and a few angels.
I dropped via the Tetra workplace to check out what Goldsmith, his co-founder Nik Liolios and one different engineer had created. Goldsmith gave me a name the use of his smartphone with the Tetra app installed. As he and the deep learning models working in the historical past listened, I threw out a barrage of challenges for the transcription service.
speakme at various speeds, throwing out numbers, startup names and other tough words did stump Tetra to a few degree — however to be reasonable, there is no AI that I haven’t broken. Given how easy Tetra is to use, I may see it being used as a backup reference or for list keeping — turn it on, ignore it and use it to go looking via notes later.
In cases where 99 or 100% accuracy is required, Tetra presents human transcription for a price and a 24-hour wait. This actually helps each shoppers and Tetra in the sense that accurate transcriptions can feed back as working towards statistics to improve future performance.
Goldsmith advised me he’s finding traction selling to traders making usual diligence calls. These shoppers need Tetra to create a everlasting listing of conversations with industry consultants. different, more average, business use circumstances exist as neatly, like inside revenue.
This looks to be working out relatively neatly for the company. And issues stay pretty lean with the three-person Tetra crew figuring out of a residential residence dually zoned for industrial. On the engineering facet, a lot of the underlying infrastructure is being powered via off-the-shelf APIs.
this is in fact a good factor, because it capacity Tetra isn’t wasting time building issues that already exist available on the market and instead is focusing on accumulating a massive transcriptions statistics set so as to best proceed to enhance the excellent of the provider relocating forward.
The crew’s method is heavily elegant on being able to optimize which parts of conversations are sent to which cloud API. for example, some NLP provider suppliers are better at understanding speech relating to movies, song and media, whereas others are better at numbers, and many others.
The $ 1.5 million in seed financing is going for use to scale up the engineering group and enrich desktop researching pipelines. Tetra comprises search functionality so clients can without delay locate selected sentences within historically unsearchable voice recordings. I might see this fitting extra proactive sooner or later — flagging names and dates immediately, as an example.
Fundings & Exits – TechCrunch