Natural Language Processing news and analysis on Web Pulse: 5424 curated articles tracking the latest Natural Language Processing developments, tools, and research, updated continuously from vetted sources.
Researchers introduced Gaussian Process Embedding Correction (GPEC), a pre-LLM error-correction method that improves cardiac ultrasound caption generation by VideoChat2, according to an arXiv paper (arXiv:2610.00196v1). …
A new arXiv paper, arXiv:2610.00791v1, introduces Enterprise Representation Simplification (ERS) and Enterprise Representation Complexity (ERC), a representation-neutral model that measures representational extent across…
A developer has published a guide on building semantic search engines in Java using vector embeddings, walking through the architecture of an embedding model, a vector search engine, and similarity ranking. The writeup c…
Microsoft Corp. launched MAI-Transcribe-2-Streaming, its first streaming transcription model, priced at $0.54 per audio hour, alongside text-to-speech models MAI-Voice-2.1 at $22 per million characters and MAI-Voice-2.1-…
A practical guide explains tokens per second (TPS) as a benchmark metric for measuring AI model processing speed, particularly for real-time NLP applications like chatbots and voice assistants. It outlines a methodology …
OpenVoiceOS added a wake word verifier plugin that identifies enrolled household speakers at the wake gate, rejecting television audio and unrecognized voices, as part of a set of updates to the open-source voice assista…
Tavus previewed Griffin-Lite on October 1, 2026, a full-duplex model for video calls that sees, hears and talks at once, joining OpenAI's GPT-Live-1 and three open-weight models from NVIDIA and Kyutai as the five models …
A developer found that their DOCX-to-Markdown converter and RAG pipeline were quoting deleted contract clauses because unaccepted Word tracked changes persist in word/document.xml as <w:del>/<w:delText> elements. One 14-…
Researcher and OnePersonAI founder Akshat Raj has published a mathematical proof and architectural specification describing "Context-Entropy Collapse" (CEC), a phenomenon in which enterprise LLMs degrade as context lengt…
A developer has published a step-by-step guide for building an AI-powered voice call application that combines Twilio's telephony webhooks with ElevenLabs' text-to-speech API and OpenAI's GPT-4o-mini for conversational r…
A developer built Mock Band, a browser-based rhythm section that lets guitarists change key, tempo or feel without taking their hands off the instrument, using React, TypeScript, WebMIDI and an AudioWorklet-based audio b…
A developer building a RAG pipeline over converted EPUB books found that invisible U+00AD soft hyphens inherited from print typography were breaking full-text search and embeddings, with 4,213 such characters counted in …
A developer outlined how Python's asyncio enables non-blocking I/O for AI agent workflows and how the Pydantic library enforces structured data validation on unstructured LLM text output. The writeup walks through synchr…
Microsoft released three MAI voice models on October 1st: MAI-Transcribe-2-Streaming, MAI-Voice-2.1 and MAI-Voice-2.1-Flash, announced in a Microsoft AI post on X. The streaming transcription model accepts continuous aud…
David AI released DAI-ASR-I18N, a multilingual benchmark for conversational speech recognition and speaker diarization built from 147 hours of unscripted conversations across 21 languages. The company evaluated eight com…
Microsoft released its first streaming transcription model, which debuted at No. 1, according to the company's announcement at microsoft.ai/news/our-first-streaming-transcription-model/. The model is Microsoft's first en…
Tavus introduced Griffin, which the company calls the first "Human Interaction Model," an AI that conducts real-time video calls by processing facial expressions, tone of voice, and gestures. In a Tavus study, 48 percent…
Microsoft's MAI speech models — microsoft/mai-voice-2.1, its lower-latency sibling microsoft/mai-voice-2.1-flash, and the streaming transcription endpoint microsoft/mai-transcribe-2-streaming — are now available through …
A small experiment using Pythia-160M traces the token "light" through six contexts across its two senses — weight and color — to measure where contextual separation emerges in hidden states and how it changes near the ou…
Andrej Karpathy published an "idea file" gist, llm-wiki.md, describing a pattern in which an AI agent writes and maintains a linked markdown knowledge base from user-supplied sources, compiling synthesis at ingest time r…