@gradientbrief.bsky.socialA new arXiv study finds that pretrained self-supervised speech models like Wav2Vec2 and HuBERT, despite training data skewed toward high-resource languages, can recognize click consonants from Khoisan languages…
#AI #SpeechRecognition #NLP #LowResourceLanguages
https://arxiv.org/abs/2606.11542
@freegardener.bsky.socialStreaming ASR looks fluent but secretly gambles on every character. A new single-pass model decides when to commit vs. what to commit—no revisions, no global #SpeechRecognition #ASR #MachineLearning #AI
https://freegardner.com/synapse/machine-that-says-it-is-listening.html
The model transcribes audio, provides word timestamps, and generates speech embeddings, operating within a single 16.9 MB file on the CPU.
@ai-news.at.thenote.appTesting Android’s On-Device Speech Recognition With Hindi-English Code-Switching
On-device speech recognition turned "dizzy" into डीसी for my Hindi-speaking mother. The one-flag fix, a crashing OnePlus, and a reject button.
@ai-ru.at.thenote.appТестирование распознавания речи на устройстве Android с переключением между хинди и английским языками
Распознавание речи на устройстве превратило «головокружение» в डीसी для моей мамы, говорящей на хинди. Исправление с одним флагом, сбой OnePlus…
@ossradarai.bsky.socialBerkeley's HuPER models phonetic perception as adaptive inference, hitting state-of-the-art English error rates with just 100 hours of training and zero-shot transfer to 95 languages. Code and models are open-sourced on GitHub.
#OpenSourceAI #NLP #SpeechRecognition
https://arxiv.org/abs/2602.01634
@martincid.comGemini 3.8 Live removes the blank face from AI customer service #CustomerService #SpeechRecognition #ArtificialIntelligence
@achronicvoice.com“In a room with normal background #noise ..my #SpeechRecognition without #HearingAids is roughly 40–50%. That means I miss about half of what’s being said and have to guess the rest from context.”: buff.ly/hiTT9jz
by @donnietownstudio.bsky.social
#conversation #deaf #disability #disabled #spoonie
@lexacom.bsky.socialSimplify your work with smart workflows, ambient voice technology, and medical speech recognition on the go.
🔎 Search Lexacom on the App Store and Google Play
#nhs #digitaltransformation #workflows #productivity #ambientAI #speechrecognition
@lexacom.bsky.socialSimplified, intuitive, and packed with new features, download our new app.
Intelligent workflows, ambient voice technology, and speech recognition on the go.
🔎 Search Lexacom on the App Store and Google Play.
#nhs #digitaltransformation #workflows #productivity #ambientAI #speechrecognition
@ai-news.at.thenote.appWhy Speech Recognition Misses Human Context: Dr. Sunday David Ubur’s Affective Architecture
Dr. Sunday Ubur's research explores emotion-aware AI captions that preserve tone, urgency, and context for deaf and hard-of-hearing users.
@ai-ru.at.thenote.appПочему распознавание речи упускает человеческий контекст: Аффективная архитектура доктора Сандея Дэвида Убура
@cryptonews-poster.bsky.socialXiaomi open-sources Xiaomi-CocktailASR-1, an industrial-grade speaker diarization LLM! Solves the "cocktail party problem," precisely identifying and transcribing target speakers in multi-speaker audio using a reference voiceprint. Big leap for speech tech! #AI #OpenSource #SpeechRecognition
@hackernoon.comDr. Sunday Ubur's research explores emotion-aware AI captions that preserve tone, urgency, and context for deaf and hard-of-hearing users. #speechrecognition
@everycivicapp.bsky.socialaeneas aeneas is a Python/C library and a set of tools to automagically synchronize audio and text (aka forced alignment) https://github.com/readbeyond/aeneas #GeneralPurposeOpenSourceTools #SpeechRecognition #CivicTech
@lexacom.bsky.socialWhat do 152,416 words in a week mean to a GP?
For Dr Rafay, it means using Lexacom across all consultations, referrals and admin - saving an hour a day.
➡️ Read the full case study and start your free trial at lexacom.co.uk/faster
#nhs #generalpractice #speechrecognition #healthtech #productivity
@flo7up.bsky.socialHeidi fine-tuned and deployed medical speech recognition for clinical documentation, using synthetic multilingual data, distributed GPU training, and scalable AWS serving. The platform supports 2.4M+ consultations weekly. #HealthcareAI #SpeechRecognition #ClinicalDocumentation
@aidailypost.comMicrosoft’s new MAI‑Transcribe‑2 slashes cost and cuts transcription time in half—outpacing OpenAI’s frontier models on call‑center audio and multilingual support. Curious how fast and cheap AI can get? Dive in. #MAITranscribe2 #SpeechRecognition #MicrosoftAI