FullduplexFullduplex/blog

Fullduplex.

An observatory for Speech-to-speech, Full-duplex, and Audio Foundation Models. We collect the papers, benchmarks, models, and datasets in one place — and we keep the reply channel open.

Much of this observatory is researched and updated automatically by AI. If you spot an error, a misattribution, or an outdated entry, please report it in our community Discord.

34
models tracked
42
benchmarks
42
datasets
// latest from the observatory
native FD17
open weights21
benchmarks42
datasets42
latest releases
fastest native STS
recently updated benchmarks
§02 · top benchmarks

Benchmarks that keep score.

The evaluations we trust for speech-to-speech, full-duplex, and audio-foundation work — sorted by most-recent update. Each row points to the maintainer's live leaderboard or repo.

  1. 01
    Full-Duplex-Bench v3
    nativelive#full-duplex
  2. 02
    Big Bench Audio
    nativelab#speech-lm
  3. 03
    Scale Voice Showdown
    nativearena#speech-lm
  4. 04
    τ³-Bench (τ-Voice)
    nativevertical#vertical-agent
  5. 05
    VocalBench
    nativelab#speech-lm
§03 · frontier datasets

Corpora at the frontier.

Freshly-minted full-duplex, dialog, and speech-LM corpora from 2024 – 26. The training set decides the ceiling — these are the ones actually pushing it.

  1. 01
    otoSpeech-full-duplex-280h
    permissiverole · dialog-interactiveinter. · high
    280 h
  2. 02
    otoSpeech-full-duplex-processed-141h
    permissiverole · dialog-interactiveinter. · high
    141 h
  3. 03
    InteractSpeech
    customrole · dialog-interactiveinter. · high
    150 h
  4. 04
    OleSpeech-IV
    non-commercialrole · dialog-interactiveinter. · medium
    100 h (open subset)
  5. 05
    MLC-SLM (Interspeech 2025)
    gatedrole · dialog-interactiveinter. · high
    1,604 h
§04 · verticals · long-form

The labs behind the directory.

The Verticals series profiles one lab, company, or maintainer per essay. Every piece has a matching row on /models, /benchmarks, or /datasets — tap an entry to open the essay, tap the chip to jump straight to the directory.

§05 · latest dispatches

Fresh off the press.

Our own field notes on speech-to-speech, full-duplex, and audio-foundation work — the primers, thresholds, architecture maps, data ceilings. Every post double-checked against the benchmarks and models cataloged on the observatory.

  1. 05Foundation before vertical#foundation
  2. v01Kyutai: the twelve-person Paris nonprofit turning open releases into shared vocabulary#verticals
  3. v03Cartesia: why AWS put a non-transformer voice AI on its own shelf#verticals
  4. v04Hume AI: the smile inside a sentence, and the nine days that clarified voice AI’s exit shape#verticals
§06 · community

You know a model we're missing? Tell us.

Fullduplex is, as the name implies, a two-way channel. New models, benchmarks, datasets — and corrections on our own — go through the community board.

§07 · subscribe

One email, every Wednesday.

New models, updated benchmarks, papers and datasets worth your time — one digest a week. No tracking pixels, no drip sequences, no dark patterns.