Natural Language Processing news and analysis on Web Pulse: 5439 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
Google's Gemini 3.8 Flash TTS scored 89.5% on the Artificial Analysis Pronunciation Robustness Benchmark, claiming the number one spot ahead of its predecessor Gemini 3.1 Flash TTS at 88.2%. The model, launched September…
OpenAI added its most advanced voice system, GPT-Live, to ChatGPT's website and mobile app on Wednesday, letting users speak to delegate multi-step tasks that require a browser and disparate apps. ChatGPT voice product l…
OpenAI's GPT-Live-1 voice model cuts turn-taking latency to 0.798 seconds, down from 1.41 seconds in GPT-Realtime-2.1, and reaches an 83.6% first-attempt completion rate on Tau3 retail, telecom and airline support tasks …
Google launched its Gemini 3.8 Flash TTS models on September 23, offering developers 30 prebuilt studio voices, a custom voice design system driven by natural language prompts, and support for up to 130 languages through…
Google released its Gemini 3.8 text-to-speech models, expanding the Gemini audio family that also includes Gemini 3.8 Live for real-time voice interactions and Gemini 3.8 Flash for coding tasks. The models let creators a…
Google introduced two new text-to-speech models in its Gemini family, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, expanding voice generation from 30 original voices to a library of 2,000+ production-ready voices …
A pure-Java implementation of NVIDIA's Parakeet automatic speech recognition model, called Parakeet.java, matches the reference parakeet.cpp engine word-for-word at F16 quantization with 0.00% WER difference on LibriSpee…
Researchers submitted SpeakerMem-R1, a speaker-centered dual-track memory system for multi-party dialogue, to arXiv on 22 Sep 2026. SpeakerMem-R1 achieved 62.33% on the publicly reported EverMemBench leaderboard from Eve…
NVIDIA released Nemotron 3 Diarization, an open-weight, 100M-parameter speaker diarization model that ranks #1 on Voice Arena's Diarization-Bench leaderboard with a 14.72% Diarization Error Rate. The model supports up to…
A developer documented four cases where pdfium and pypdf extract different text from the same PDF, including zero horizontal scaling (0 Tz) that pdfium ignores but pypdf reads, /ActualText replacement strings, /ToUnicode…
Tokenization converts text into byte sequences that are assigned integer token IDs before entering a transformer model, according to a technical guide on LLM text processing. The guide explains that Byte Pair Encoding (B…
Alibaba's Qwen team released Qwen-Audio-3.1, a lineup of five models spanning speech recognition (ASR), text-to-speech (TTS), and real-time interaction, and cut AI audio prices by up to 95 percent. The ASR model improves…
Alibaba's Qwen team released Qwen-Audio-3.1-TTS, a production-oriented speech synthesis system that ranks first on the independent Artificial Analysis Text-to-Speech Leaderboard. The model pairs a 12.5 Hz low-frame-rate …
Cosine similarity measures how alike two vectors are by dividing their dot product by the length of both vectors, cancelling out magnitude and leaving only the angle between them, according to the third installment of a …
A study of 3,600 exact-rational word problems and 8,600 prompts across five open-weight language models found canonical accuracy of 0.969-0.996 but orbit correctness of only 0.848-0.981 and orbit invariance of 0.851-0.98…
A new arXiv paper (arXiv:2609.25012v1) reports that diachronic word embeddings can track semantic change in Sanskrit, an ancient low-resource language, with 19 of 21 testable shifts moving in the philologically attested …
A new arXiv paper (arXiv:2609.25049v1) proposes Semantic Routing Calibration (SRC), a lightweight, training-free inference framework that reduces over-refusal in safety-aligned large language models by localizing and dyn…
Researchers demonstrated the first cross-model handoff of persistent recurrent inference state between differently sized hybrid language models without target prefix replay, transferring live memory from a Qwen3.5 4B mod…
The ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains drew 20 valid submissions from 8 teams, testing visual question answering over documents spanning eight domains including business repo…
A new arXiv paper (arXiv:2609.25056v1) proposes a graph-based framework for the Jotto word deduction problem, representing valid words as nodes in a weighted graph where edge weights equal the number of common letters be…