60
60
Measuring benchmark optimization in speech recognition
Hugging Face examines how much speech recognition systems overfit benchmarks and how to measure benchmark optimization in ASR.
A Hugging Face post on measuring benchmark optimization in automatic speech recognition, analyzing how model improvements on benchmarks reflect genuine capability gains versus overfitting. It is evaluation methodology research with no direct security impact.
22
30
45
60
30
60
30
57
45
60
60
45
Gemini Live audio
Google released Gemini 3.8 Live speech-to-speech models; Simon Willison built a library-free browser web UI for real-time voice conversations with them.
Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new speech-to-speech models comparable in shape to OpenAI's GPT-Live family. Simon Willison had GPT-6 Astra Extra High build him a web UI for testing the models, supporting model and voice preset selection, optional system prompts, and interrupting the model mid-speech. The implementation uses no libraries and talks directly to Google's wss://generativelanguage.googleapis.com BidiGenerateContent WebSocket endpoint.
45
Google launches Gemini 3.8 Live to take on OpenAI's GPT-Live-1 at a fraction of the cost
Google DeepMind launches Gemini 3.8 Live speech-to-speech models, topping Artificial Analysis leaderboard at 82.6% with much cheaper pricing than OpenAI.
Google DeepMind released Gemini 3.8 Live and 3.8 Live Extended Thinking, audio models for voice agents available through the Gemini API and Google AI Studio, supporting over 97 languages plus background API calls and visual input. The Extended Thinking variant ranks first on the Artificial Analysis Speech-to-Speech Leaderboard with 82.6%, ahead of OpenAI's GPT-Live-1 models. Google charges $0.005 per minute for audio input and $0.018 for output, versus OpenAI's $0.05 per minute, though OpenAI retains full-duplex conversation quality advantages.
55
The Semantic Bottleneck: Leveraging Semantic Representations for Non-Invasive Speech Decoding
Researchers introduced Brain2Semantics2Text, decoding sentence meaning from non-invasive MEG brain recordings via a semantic bottleneck, improving on prior Brain2Text methods.
The paper proposes Brain2Semantics2Text, a non-invasive speech decoding method that maps sentence-level magnetoencephalography (MEG) responses into a semantic embedding space and inverts those embeddings into natural language. Motivated by evidence that high-level semantic representations are distributed across cortex and evolve on slower timescales, the approach targets meaning rather than phonemes or words, avoiding the need for word-level alignment. The authors report improved sentence-level results compared to prior non-invasive Brain2Text methods despite the low signal-to-noise ratio of neural recordings.
25
Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech
Researchers distill a compact 82M-parameter Thai TTS from synthetic OmniVoice data, enabling on-device fixed-voice synthesis without reference audio.
The paper uses a large voice-cloning model (OmniVoice) as a synthetic data source to train Wayu-Paxa-TTS-Edge, an 82M-parameter fixed-voice Thai TTS student. The model achieves 68.2% Challenge-Set Keyword Accuracy (85.5% of Gemini 3.1), 91.4% pause precision, and CERs of 3.7% on Thai and 1.1% on English. It outperforms its teacher on pause placement and is open-sourced with its evaluation framework.
25
30
30
30
30
55
30
55
42
60
55
45
60
45
30
60
30
60
45
Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents
Duplex Cue evaluation shows PersonaPlex full-duplex agents adapt in-turn to listener contributions in only 34.8% of collaborative cases versus 68.2% for humans.
The paper introduces Duplex Cue, an evaluation of in-turn adaptation in full-duplex voice agents that separates listener intent (backchannel, collaboration, interruption) from speaker behavior (continue, adapt, yield). Using 208 scorable pairs from 300 human-confirmed cues in unscripted English conversations, it compares recorded human responses with PersonaPlex continuations generated while replaying listener audio. Humans adapt within the turn in 68.2% of collaborative pairs versus 34.8% for PersonaPlex, which otherwise continues unchanged (42.4%) or yields (22.7%).
15
Gradium Launches Voice Design: Write a Prompt, Get a Brand New Synthetic Voice in Seconds
Gradium, a Kyutai spinout, launched Voice Design, generating custom synthetic voices from text descriptions in seconds across five languages.
Gradium, a Paris-based voice AI company spun out of Kyutai, launched Voice Design, which generates new synthetic voices from 1-500 character text descriptions in seconds without needing reference audio or speaker consent. The feature is live in the Gradium API and Studio, free on every plan including the free tier, and kept voices run on the standard streaming TTS endpoint at the same latency as catalog voices. Vendor-run blind pairwise listening tests across 7,627 comparisons report a 72.6% win rate, 13.6 points ahead of ElevenLabs at 59.0%, placing first in all five tested languages, with the largest margins on regional accents such as Quebecois French (97%).
32
BreezeBlue/Breeze-TTS-2 — new model trending #19 on Hugging Face
BreezeBlue open-weights Breeze TTS 2, a bilingual text-to-speech model it ranks #1 among open-weight models on the Artificial Analysis TTS leaderboard.
BreezeBlue released open weights and Apache 2.0-licensed PyTorch inference code for Breeze TTS 2 on 2026-08-25. The text-to-speech model supports English and Chinese, voice cloning, reference-free voice design, voice direction, and inline vocal events like (laugh) and (sigh). Reported performance includes #1 open-weight ranking on the Artificial Analysis Elo leaderboard, under 40 ms time-to-first-audio, a 0.32 real-time factor on an NVIDIA H100, and about 7.7 GiB GPU memory for eager inference.
45
Bad Likert Judge: A Novel Multi-Turn Technique to Jailbreak LLMs by Misusing Their Evaluation Capability
Unit 42 details the Bad Likert Judge multi-turn jailbreak that abuses LLMs' evaluation capability, raising attack success rates over 60% across six frontier models.
Palo Alto Networks Unit 42 describes the Bad Likert Judge technique, a multi-turn jailbreak that asks a target LLM to act as a Likert-scale judge scoring the harmfulness of example responses. The highest-rated example in each scale can carry harmful content, bypassing the model's internal guardrails. Testing across six state-of-the-art text-generation LLMs showed an average attack success rate increase of more than 60% versus plain attack prompts, with tested models anonymized. The technique targets edge cases rather than typical use, and the article positions the work as guidance for defenders on potential jailbreak risks.
60