The dominant architecture pattern for text-to-speech in 2026 looks a lot like an LLM — an autoregressive transformer generating sequences of tokens, one frame of audio at a time. Samuel Humeau from Mistral walks through why the field converged there, how neural audio codecs solve the information-density problem (audio carries ~200kbps of signal; you can't feed that raw to a transformer), and what the streaming trick actually is that makes voice agents feel responsive before the full audio has even finished generating.

The talk uses Mistral's just-released open-weight TTS model as a running example — live demos of voice cloning from a few seconds of reference audio, a voice agent answering real conference schedule questions, and a breakdown of the codec-to-backbone-to-decoder pipeline that produces it all. There's also a frank section on what's still unsettled: how to handle streaming text input (tokens arriving from an LLM in real time rather than a fixed block of text) and why getting that right is the next meaningful latency win in agent pipelines.

It's the kind of talk that makes the system feel less like a black box — not by oversimplifying, but by showing exactly which engineering choices are load-bearing and which are still open problems.

Speaker info:

https://blenderdumbass.org/games/Dani's_Race

Reverse eng version of CL MMORPG that got canceled.

The "Her" moment has been promised so many times it's become a joke. Every new demo, every smooth-sounding voice agent gets called it. Neil Zeghidour, CEO of Gradium AI and one of the researchers behind Moshi — the first full-duplex voice model — uses this talk to be honest about where the gap actually is and why it keeps not closing.

The core tension: cascaded systems (speech-to-text, LLM, text-to-speech) are practical and getting smarter, but they're architecturally incapable of feeling like a real conversation. Latency from tool calls alone can be 500ms to 4 seconds — while humans process and respond in around 200ms total. Speech-to-speech models solve some of that but trade it for a different problem: they're still half-duplex, meaning they're either listening or talking but never both, which makes backchanneling impossible and the interaction feel robotic in a different way. Moshi showed that full-duplex is solvable. What it didn't solve was making the model useful. And cost is a wall hiding behind the latency problem — TTS at scale is expensive enough that some teams burn through their fundraising before they can grow a user base.

The most underrated thread in the talk is paralinguistic understanding: voice carries tone, hesitation, discomfort, and cultural signals that get entirely stripped out the moment you transcribe to text. Getting to Her means building models that don't just produce natural-sounding speech but actually understand what the voice is carrying — and that's a science problem, not a prompt engineering one.

Speaker info:

Un regard impertinent et libre, orchestré par Patrick Menais et son équipe, sur le monde de l’image.

En avant-première du lundi au samedi à 17h00 sur Facebook, Twitter et YouTube.

Du lundi au samedi à 20h00 sur France 5.

https://www.facebook.com/vufrancetv

Chat agents dominated 2025. Every product either went AI-first or got left behind. But text-in, text-out is already starting to feel dated. Voice is faster, more accessible, and opens up interaction paradigms that chat just can't touch — phone lines, Zoom calls, screen readers, ambient interfaces. In this talk, Luke Harries from ElevenLabs argues that the next upgrade for every chat agent isn't better prompts or smarter RAG. It's a voice layer.

The problem is most teams have already built and tuned their chat agents. They don't want to throw that out. This session shows how ElevenLabs' Voice Engine wraps any existing agent in a few lines — handling turn-taking, speech-to-text, text-to-speech, and emotion-aware interruption detection — without touching the underlying logic. There's a live demo of converting a working chat support agent to voice in a single prompt, plus a look at the client and server SDKs, Shadcn-based UI components, and how tool calling still works through the wrapper.

Speaker info:

COMMANDES UTILES CÔTÉ SERVEUR

Mise à jour du système :

Installation du paquet serveur NFS :

Édition du fichier "exports" :

Ajoutez la ligne suivante (exemple à adapter selon votre configuration) :

Prise en compte des modifications :

Vérification de l'export :

Si l'export n'est pas effectif, redémarrez le service :

COMMANDES UTILES CÔTÉ CLIENT

Installation du paquet client NFS :

Pour lister les partages disponibles sur un serveur (remplacez "serveur.local" par l'IP du serveur) :

Création du point de montage (le dossier où apparaîtront vos fichiers) :

Configuration du montage automatique via le fichier fstab :

Ajoutez cette ligne au fichier (à adapter selon vos paramètres d'IP et de chemins) :

LIENS UTILES

👉 Network File System sur Wikipédia

Bon, et bien le PBS (Proxmox Backup Server) dédié attendra... car la machine qui devait lui être consacrée s'est finalement transformée en serveur NFS 😉

Note importante : En début de vidéo, je mentionne un RAID1, puis un RAID0. C'est bien de RAID0 dont il s'agit pour cette configuration ! 😊

De la cuisine au garage, en passant par le bureau et mon fameux fauteuil jaune moutarde délavé, je vous emmène pour la mise en place de ce serveur dédié NFS.

Dans la prochaine vidéo, je vous montrerai les étapes détaillées pour créer votre propre espace de stockage NFS et, bien sûr, comment y accéder depuis un poste client sous Linux ou Windows 👍

Installé dans mon fauteuil jaune moutarde délavé (😜), je teste rapidement Warpinator sur mobile Android 😮

Si vous voulez en savoir plus, voici la traduction de l'article de blog de Linux Mint. On y retrouve, bien entendu, l'annonce officielle de la version mobile de Warpinator 👍

LIEN UTILE

👉 Linux Mint : Nouvelles mensuelles d'avril 2021

Le thé, boisson parmi les plus consommées au monde, vient de Chine mais existe en de nombreuses variétés selon les continents. "Le dessous des cartes" décrypte la mondialisation et l’industrialisation du thé autour de quelques géants du secteur. Quels sont les enjeux environnementaux liés à cette industrie, notamment avec l’engouement mondial pour le thé matcha?

Magazine (France, 2026, 12mn)

🎓 Découvrez l'Académie du Dessous des cartes 👉 https://www.academiedudessousdescarte...

👉 https://www.arte.tv/fr/newsletter/ddc/

📕 Commandez le nouvel atlas "Le Dessous des Cartes - La puissance et la mer", d'Émilie Aubry et Frank Tétart, aux éditions Tallandier et Arte Editions https://bit.ly/Atlas_DDC

👉 Abonnez-vous à notre chaîne ! http://bit.ly/DDC_YouTube

👉 Retrouvez également Le Dessous des cartes sur Facebook. / ledessousdes. .

💻 Retrouvez toutes les émissions d'Émilie Aubry en VOD : http://bit.ly/DDC_VOD.

👉 L’intégrale des émissions de Jean-Christophe Victor (VOD et DVD) sur la boutique d’ARTE. http://bit.ly/DDC_Boutique