Deepgram has made a reputation for itself as one of many go-to startups for voice recognition. In the present day, the well-funded firm introduced the launch of Aura, its new real-time text-to-speech API. Aura combines extremely practical voice fashions with a low-latency API to permit builders to construct real-time, conversational AI brokers. Backed by massive language fashions (LLMs), these brokers can then stand in for customer support brokers in name facilities and different customer-facing conditions.
As Deepgram co-founder and CEO Scott Stephenson informed me, it’s lengthy been doable to get entry to nice voice fashions, however these have been costly and took a very long time to compute. In the meantime, low latency fashions are likely to sound robotic. Deepgram’s Aura combines human-like voice fashions that render extraordinarily quick (sometimes in nicely below half a second) and, as Stephenson famous repeatedly, does so at a low value.
Picture Credit: Deepgram
“Everybody now is like: ‘hey, we need real-time voice AI bots that can perceive what is being said and that can understand and generate a response — and then they can speak back,’” he mentioned. In his view, it takes a mixture of accuracy (which he described as desk stakes for a service like this), low latency and acceptable prices to make a product like this worthwhile for companies, particularly when mixed with the comparatively excessive value of accessing LLMs.
Deepgram argues that Aura’s pricing at present beats nearly all its rivals at $0.015 per 1,000 characters. That’s not all that far off Google’s pricing for its WaveNet voices at 0.016 per 1,000 characters and Amazon’s Polly’s Neural voices on the identical $0.016 per 1,000 characters, however — granted — it’s cheaper. Amazon’s highest tier, although, is considerably costlier.
“You have to hit a really good price point across all [segments], but then you have to also have amazing latencies, speed — and then amazing accuracy as well. So it’s a really hard thing to hit,” Stephenson mentioned about Deepgram common strategy to constructing its product. “But this is what we focused on from the beginning and this is why we built for four years before we released anything because we were building the underlying infrastructure to make that real.”
Aura gives round a dozen voice fashions at this level, all of which have been educated by a dataset Deepgram created along with voice actors. The Aura mannequin, identical to the entire firm’s different fashions, have been educated in-house. Here’s what that feels like:
You possibly can strive a demo of Aura right here. I’ve been testing it for a bit and although you’ll generally come throughout some odd pronunciations, the pace is basically what stands out, along with Deepgram’s current high-quality speech-to-text mannequin. To spotlight the pace at which it generates responses, Deepgram notes the time it took the mannequin to begin talking (usually lower than 0.3 seconds) and the way lengthy it took the LLM to complete producing its response (which is often slightly below a second).
