EngineeringPrimary topic: AI AgentsAlso: NLP & Speech
Crest summary (AI-written):Google published a step-by-step guide for building a voice agent with its open-source Agent Development Kit (ADK) and the gemini-3.8-live model.
Companies: Google
LaunchTopic: NLP & Speech
Crest summary (AI-written):Jack Dorsey's Block announced an AI model called Dessa, which it said was trained from scratch on human conversation.
Companies: BlockJack Dorsey
LaunchTopic: NLP & Speech
Crest summary (AI-written):Dessa released a full-duplex speech model that it describes as the most authentic ever built.
LaunchTopic: NLP & Speech
Crest summary (AI-written):Microsoft introduced a voice AI that begins working out what a user wants before they finish speaking.
Companies: Microsoft
LaunchTopic: NLP & Speech
Crest summary (AI-written):IBM released Granite Speech 5.0, a speech model that transcribes, translates and attributes speech more accurately and faster than its predecessor.
Companies: Granite
EarningsTopic: NLP & Speech
Crest summary (AI-written):ElevenLabs has doubled its valuation to $22bn through a tender offer, according to Sifted.
Companies: ElevenLabs
LaunchTopic: NLP & Speech
Crest summary (AI-written):Hugging Face launched the Open TTS Leaderboard, a scalable evaluation system for multilingual text-to-speech and voice cloning models.
Companies: Hugging Face
LaunchTopic: NLP & Speech
Crest summary (AI-written):Google DeepMind released Gemini 3.8, a text-to-speech model, on 2026-09-23.
Companies: Google DeepMind
OtherPrimary topic: AI AgentsAlso: NLP & Speech
Crest summary (AI-written):Ringg's AI agents, built on OpenAI's GPT-5.6, resolve up to 65% of customer calls, OpenAI said on 2026-09-23.
LaunchPrimary topic: NLP & SpeechAlso: LLM Infrastructure
Crest summary (AI-written):Google's Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS text-to-speech models are now available on Vercel's AI Gateway, generating speech from text in over 100 languages.
LaunchPrimary topic: NLP & SpeechAlso: ML Frameworks & Training
Crest summary (AI-written):Hugging Face released tokenizers v1, a major update to its text tokenization library, with encode and decode performance and scaling measured in the announcement.
Companies: Hugging Face
LaunchPrimary topic: NLP & SpeechAlso: LLM InfrastructureAlso: API & Integration
Crest summary (AI-written):OpenAI's GPT-Live 1, a full-duplex voice model that can listen and speak simultaneously, is now available on Vercel's AI Gateway. It removes the turn-detection boundary found in many voice models.
LaunchPrimary topic: NLP & SpeechAlso: LLM Infrastructure
Crest summary (AI-written):Google's Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking models are now offered on Vercel's AI Gateway, both supporting real-time spoken interactions.
Companies: GoogleGoogle DeepMindVercel
LaunchPrimary topic: NLP & SpeechAlso: API & Integration
Crest summary (AI-written):OpenAI released GPT-Live-1 in its API, a model for building more natural voice experiences.
Companies: OpenAI
LaunchPrimary topic: NLP & SpeechAlso: Computer Vision
Companies: Hugging Face
UpdateTopic: NLP & Speech
Crest summary (AI-written):Hugging Face's Open ASR Leaderboard added its first Global South language, expanding the benchmark beyond its previous language coverage.
Companies: Hugging Face
LaunchPrimary topic: NLP & SpeechAlso: API & Integration
Crest summary (AI-written):Google's Gemini 3.5 Transcribe is now available on Vercel's AI Gateway, handling both recorded and live audio. Two model identifiers are offered: one for full-file transcription and one for live audio.
Companies: GoogleGoogle DeepMindVercel
ResearchTopic: NLP & Speech
Crest summary (AI-written):A Hugging Face study measures how much speech recognition benchmark scores improve when models are tuned directly on test data, finding gains that do not reflect real-world performance.
Companies: Hugging Face
LaunchPrimary topic: NLP & SpeechAlso: LLM Infrastructure
Crest summary (AI-written):Fish Audio's audio models are now available on Vercel AI Gateway, and every Fish Audio model is free there for 30 days through September 18.
Companies: Fish AudioVercel
LaunchPrimary topic: NLP & SpeechAlso: AI Agents
Crest summary (AI-written):NVIDIA released Magpie TTS open weights on Hugging Face for building low-latency multilingual voice agents with full deployment control.
Companies: Hugging FaceNvidia
EngineeringTopic: NLP & Speech
Crest summary (AI-written):OpenAI described building GPT-Live, a realtime system for responsive voice AI, in six months.
Companies: OpenAI
LaunchPrimary topic: AI AgentsAlso: NLP & Speech
Crest summary (AI-written):avatarin has built a 24/7 retail agent using OpenAI's GPT-Realtime, according to an OpenAI post dated 2026-07-30.
LaunchPrimary topic: NLP & SpeechAlso: LLM Infrastructure
Crest summary (AI-written):xAI's Grok Voice Think Fast 2.0, a speech-to-speech model that accepts audio input and returns audio output, is now available on Vercel's AI Gateway.
UpdatePrimary topic: NLP & SpeechAlso: API & Integration
Crest summary (AI-written):Vercel's AI Gateway added streaming transcription, letting audio be sent as it is captured and transcript updates returned as the model processes it, instead of requiring a complete audio file.
Companies: Vercel
OtherTopic: NLP & Speech
Crest summary (AI-written):Cars24 is using OpenAI-powered voice technology to scale conversations and build faster, according to an OpenAI customer story.
