- Sonos adds AI talk improvement option to ARC Ultra
- It’s Sonos’ first AI sound function with four levels of speech boost
- It is developed with a hearing organization to help people with hearing loss
Sonos has launched a new version of its speech improvement tools for Sonos Arc Ultra, which we consider one of the best available sound beams.
You will still find these tools on the now played screen in the Sonos app, but instead of having just a few options, you now have four new modes (low, medium, high and max), all driven by the company’s first use of an AI audio treatment tool. They must be available today (May 12) for all users.
These states were developed in a year -long partnership with the Royal National Institute for the Deaf (RNID), Britain’s leading charity for people with hearing loss. I talked to Sonos and Rnid to get the interior story about its development here – but you can read on here for several of the details.
The update is launched today at Sonos Arc Ultra Soundbars, but will not be available on any other Sonos Soundbars because it requires a higher level of treatment effect that the chip inside Arc Ultra can give, but the older sound beams can’t.
The AI element is used to analyze the sound passing through the sound beam in real time, separating the ‘speech’ elements from the sound so that they can be made more prominent in the mixture without affecting the rest of the sound too much. I’ve heard it in action during a demo at Sonos’ British product development facility, and it’s very impressive.
If you have used speech improvement tools before, you are probably familiar with hearing the dynamic range of sound, and especially the base, suddenly massively reduced in exchange for the speech elements being pushed more forward.
This is not the case with Sonos’ new condition – powerful bass, the overall soundscape and the more immersive Dolby Atmos elements are maintained all much better. It is for two reasons: One is that the speech is improved separately to other parts, and the other is that it is a dynamic system that only activates when it detects that speaking is likely to be drowned by background noise.
It is not activated if dialogue occurs on a quiet background or if there is no dialogue in the stage. And it is a system that works in degrees – it uses more treatment in the busiest scenes, and less when the sound is not so chaotic.
How does it sound?
In the two lowest conditions, the dialogue is chosen more clearly without any major damage to the rest of the soundtrack, based on my demo.
In the high state, the background was still maintained really well, but the speech started to sound a little more processed, and at Max I could hear the background getting its wings cut a little, and a little more artificiality to the speech – but the speech was extremely well chosen and this condition is only really designed for the hard.
I mentioned that the condition was developed with Rnid, who involved Sonos Consulting with audio research experts on RNID, but also to get people with different types and levels of hearing loss to test conditions at different stages of development and give feedback.
I spoke in the long run with Sonos Audio and AI architects that developed the new states, as well as RNID, but the most important takeaway is that the collaboration led to Sonos, which attaches greater emphasis to preserving the immersive sound effects and adding four levels of improvement instead of the originally planned three.
Despite RNID’s commitment, the new state is not designed to be exclusively to hear. It’s still just called speech improvement as it is now, and it’s not hidden away as an accessibility tool – sound improves for everyone, and ‘everyone’ now includes better people with mildly to moderate hearing loss. The low and medium -sized conditions can also just work for those of us who need some extra clarity in busy scenes.
This is not the first use of AI-driven speech separation, I have seen-I have experienced it on Samsung TVs and in a funny showcase from Philips TVs, where it was used to disable the comment during sports, but preserve the audience sound.
@Techradar ♬ Original Sound – Techradar
But it is interesting that this is the first use of AI audio treatment from Sonos, and the four-year development process, including a year of refinement with RNID, shows that Sonos has taken a thought-provoking approach to how best used that is not always visible in other AI audio processing applications. Here is my piece that interviews Sonos’ AI and audio developers with researchers from Rnid.
It’s just a shame that it’s exclusive to Sonos Arc Ultra for now – though I’m sure new versions of Sonos Ray and Sonos Beam Gen 2 will be together before too long with the same upgraded chip to support the feature.
