frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: Sparrow-2 – Solving the cocktail party problem

https://www.tavus.io/blog/sparrow-2
6•code_brian•58m ago
Hey there, I’m Brian. I've been shipping conversational models over here at Tavus for the past two years. I want to tell you about our new audio-understanding model: Sparrow-2! It’s a new category of model and a unique new approach to conversational audio.

Earlier this year we launched Sparrow-1, (at the time) our SoTA turn taking model. Since our Sparrow-1 launch, I’ve spent a lot of time listening to humans talking and trying to really understand how people know when to talk, when to listen, and when to wait. I’ve also been hunting down failure modes of the current SoTA models. And while Sparrow-1 is great, there are some failure patterns I see. We tried solving the problems with existing approaches, but solving one problem only created another.

Sparrow-1 and a lot of the existing turn taking models require noise cancellation to isolate the speaker’s audio from background noise. After removing the “noise” these models rely on simple prosodic and phonetic cues from spoken words to decide when a turn has ended. They throw out information and then pay attention to a small set of verbal and prosodic cues. This approach ignores a ton of important information.

Non-verbal cues, sounds, and environmental noise impact turn taking. Noise cancellation models assume everything non-transcribable is noise, but that’s wrong. Humans use breath, sighs, and other sounds to hold the conversational floor or bid for a turn. There are constant micro-interruptions, affirmations, quiet human and environmental sounds that add to the conversational scenario. That sound is part of what we use to understand the nuanced space of the conversational floor! Cancelling out the “noise” has the ruinous side-effect of cancelling out the flow.

Now, with Sparrow-2, instead of modelling just the primary speaker’s transcribable audio, we’ve been training a model on all the sounds and letting it decide what matters and what does not when it comes to conversational flow. Our new model continuously streams in audio and is able to understand semantics, prosody, timing, speaker identity, sighs, breaths, backchannels, interruptions, background speech, and unintelligible audio with the goal of understanding what the agent should do given the state of the audio.

Sparrow-2 fits into our larger model family and unlocks new conversational behavior, not seen before in production-ready conversational pipelines. It can pass signals to different parts of the conversation. For example if the user is in a room that’s too noisy, the bot will ask the user to move to a quieter space. Also, Sparrow-2 is semi-duplex: it considers the timing of sound in relation to the AI speech as well as the user’s.

Our main objective with Sparrow-2 is to finally crack the Cocktail Party Problem: how can we have a human-like conversation 1:1 with a user in noisy environments.

I wrote up some technical details about the model architecture here: https://www.tavus.io/blog/sparrow-2

I’d love for you guys to give it a try and let me know what you think and feel.

London: Drawn Like a Painting

https://sheets.works/data-viz/london
1•lordbubbles•2m ago•0 comments

How to Deploy Your Model

https://htdym.sailresearch.com/
1•bobstax•4m ago•0 comments

Anthropic Announces Hardware Interface Standard (Model Hardware Standard, MHS)

https://twitter.com/AnthropicAI/status/2093038426140651791
2•zof3•6m ago•0 comments

History of the Atlantic Cable and Undersea Communications

https://atlantic-cable.com/
1•CaesarA•6m ago•0 comments

Poll: Are You Aphantastic

1•andrewla•7m ago•1 comments

A Training Cluster for This Economy

https://derekrodriguez.dev/a-training-cluster-for-this-economy/
1•dwrodri•8m ago•0 comments

Don't use a free iPhone that just turns up at your home

https://appleinsider.com/articles/26/08/27/psa-dont-use-a-free-iphone-that-just-turns-up-at-your-...
2•speckx•10m ago•0 comments

Show HN: Scrub Every Single Mention of Agents from the Browser

1•aihatterer•11m ago•0 comments

CIA Has a Gen Z Problem – Young Spies Are Just Not 'Feeling It'

https://spyscape.com/article/the-cia-has-a-gen-z-problem-young-spies-are-just-not-feeling-it
2•jumpocelot•13m ago•0 comments

Trump signs order to rename Lake Ontario as Lake America

https://www.reuters.com/world/us/trump-signs-order-rename-lake-ontario-lake-america-2026-08-27/
3•cdrnsf•13m ago•3 comments

ClickHouse winning this observability "war"?

https://clickhouse.com/blog/is-clickhouse-winning-the-observability-wars
1•gingenio•15m ago•0 comments

Nvidia Starts Pac as AI Chip Maker Builds DC Influence Force

https://news.bgov.com/bloomberg-government-news/nvidia-starts-a-pac-as-ai-chip-maker-buids-influe...
2•rarisma•16m ago•0 comments

Danrh: A blog won't give you an edge anymore

https://danrh.net/posts/2026-05-18-a-blog-is-not-enough/
1•speckx•16m ago•0 comments

Blogging with (Actual) Paper

https://jhuleatt.com/posts/blogging-with-actual-paper/
1•jhuleatt•16m ago•0 comments

Show HN: Husk a Local-first, open-source, security scanner

https://husk-security.dev/
1•PatrikTang•17m ago•0 comments

The Caped Billboard Bandit of Chicago

https://slate.com/life/2026/07/chicago-superheroes-smoking-cigarettes-black-americans.html
1•bryanrasmussen•17m ago•0 comments

In Hilbert Space, All Things Are Quantumly Possible – Charlie Wood, Quanta

https://www.quantamagazine.org/in-hilbert-space-all-things-are-quantumly-possible-20260826/
1•unprovable•17m ago•0 comments

Show HN: Easily deploy vibecode apps through MCP

https://zerolith.io/
1•yojukawa•18m ago•0 comments

A Fetishist's Guide to the Monochrom (Part 1) (2012)

https://leica-camera.blog/2012/10/23/fegor-a-fetishists-guide-to-the-monochrom-part1/
1•multisport•20m ago•0 comments

Show HN: Z, minimal agentic harness for engineers that just works

https://github.com/computerex/z
1•computerex•22m ago•0 comments

Aether – live visual su Mac. Create what you hear

https://centroscs.it/pages/aether.php
1•mkovideolab•23m ago•1 comments

Y Combinator's Ankit Gupta faces backlash over H-1B comments

https://americanbazaaronline.com/2026/08/27/y-combinators-ankit-gupta-faces-backlash-over-h-1b-co...
4•Tabular-Iceberg•25m ago•0 comments

Temporal at Loop

https://www.loop.com/engineering-blog/temporal-at-loop---evolution-and-scaling
1•vzhou842•25m ago•0 comments

The AI 'Ghosts' Contaminating Academic Publishing

https://www.404media.co/the-ai-ghosts-contaminating-academic-publishing/
1•saikatsg•25m ago•0 comments

Show HN: Memorial for Athena Strand. She will always be remembered

https://athena-strand.org
2•computerex•27m ago•1 comments

Autism mutations drive neurodevelopmental pathology

https://www.science.org/doi/10.1126/science.ady4523
3•slantedview•27m ago•0 comments

Show HN: Collaborate with your team on Claude and ChatGPT artifacts

https://www.valet.dev/blog/your-agent-is-only-the-first-collaborator
3•miradu•28m ago•0 comments

Launch of Biological Data Center Prototype in Singapore

https://medicine.nus.edu.sg/news/nus-medicine-dayone-and-cortical-labs-unveil-biological-data-cen...
2•tow21•29m ago•0 comments

Data Became Code: We Ran Code Inside Fortune 500s

https://medium.com/@alonhertz1/data-became-code-we-ran-code-inside-fortune-500s-using-files-they-...
2•jonbaer•30m ago•0 comments

Pushing GPT-5.6 Luna from 0% to 56% on ARC-AGI-3 Public

https://int21.ai/insights/pushing-gpt-5-6-sol-to-100-on-arc-agi-3-public/
3•antinucleon•31m ago•0 comments