Speech to text is the automated transcription of spoken audio into written text using machine learning models. Truncated covers research developments, transcription accuracy benchmarks, software library releases, and hardware acceleration.
AI-generated summary Google is expanding its Gboard voice dictation feature known as Rambler to the Pixel 10 series. The November Pixel Drop also brings other Gemini Intelligence features to the Pixel 10 lineup.
AI-generated summary ElevenLabs is launching two new Speech to Text models named Scribe v3 and Scribe v3 Live to improve transcription accuracy and handle messy real-world audio. ElevenLabs says Scribe v3 makes 17 percent fewer diarization errors than Scribe v2 on recordings with three or more speakers.
AI-generated summary Cartesia introduces Joyce as a researcher working on the speech-to-speech team. Cartesia shares a personal story about the researcher through a published link.
AI-generated summary Google details on-device speech recognition and text-to-speech features available in the Google AI Edge Gallery for developers. The open-source code on GitHub provides optimized solutions using ML Kit, MediaPipe Tasks, and LiteRT for Android and cross-platform applications.
AI-generated summary OpenRouter Blog published a tutorial on building a voice AI restaurant booking agent in Python that transcribes callers, uses tools to check availability, and converts responses into speech. OpenRouter Blog states that across 12 test turns, the median server-side time to first audio was 1.9 seconds, with text-to-speech accounting for roughly 90% of the measured model cost.
AI-generated summary The announcement introduces MAI-Transcribe-2 as a transcription model priced at $0.10 per hour of audio. The publisher states that the model is 10x faster than GPT-Transcribe and 5x faster than Gemini 3.5.
AI-generated summary An internal test build of the Samsung Keyboard app includes a new option called Refine Voice Input powered by Galaxy AI. The reported feature is designed to clean up filler words and hesitations during voice dictation on supported Galaxy phones.
AI-generated summary The Federal Bureau of Investigation reports the adoption of Superintelligence programs and tools like Microsoft CoPilot Chat, Hippo Chat, and AJae to automate transcription, translation, and data review. Director Kash Patel states that these technology upgrades and Digital Automation Solutions have saved hundreds of thousands of labor hours across field offices and investigations.
AI-generated summary Otter.ai announced the expansion of its multilingual capabilities to support 31 languages in total across global meetings. The updated tool introduces a feature that follows conversations dynamically as speakers switch languages during the same call.
AI-generated summary Anthropic reports that dictation is now available for all Claude users across Free, Pro, Max, Ultra, Team, and Enterprise plans on iOS and Android devices. Anthropic explains that the dictation feature turns spoken speech into text prompts while support for languages other than English remains in beta.
AI-generated summary NYU Langone Health announced that its PREVIA ensemble analyzed emergency-department trauma narratives to estimate one-month post-traumatic stress disorder risk in a 145-patient study. The system utilizes Whisper, Llama, Mistral, DeepSeek, and NVIDIA A100 GPUs to process patient recordings and demographic data without adding intake time.
AI-generated summary Boson AI co-founder and CEO Alex Smola says the startup is expanding its Toronto office faster than its Silicon Valley headquarters to build voice AI foundation models. Smola attributes the Canadian expansion to predictable conditions, easier work permit acquisition for foreign talent, and partnerships with local institutions like the University of Toronto.
AI-generated summary MistralAI used Voxtral 3 Transcribe at AI Engineer Paris to capture every main-stage talk in real time. MistralAI boosted accuracy with context biasing by feeding the model speaker names and technical terms from the conference agenda.
AI-generated summary AppleInsider reports that the developer beta of iOS 27.3 introduced a new Apple Intelligence feature under its Audio Intelligence banner. Using an Apple Watch Ultra 4 or Apple Watch Series 12, the wearables will listen to conversations and transcribe 15 seconds at a time.
AI-generated summary Popular Science reports that artificial intelligence advances enable numerous applications to recognize human speech and generate transcriptions of video and audio meetings. Users can utilize built-in tools across Zoom, Google Meet, and Microsoft Teams, or choose third-party options like Scribbl, Otter.ai, Buzz Captions, Spokenly, and MacWhisper.
AI-generated summary Raycast released a new Dictate to AI command that allows users to speak requests while automatically providing screen context from the active app. The latest release also updates Raycast Notes, improves performance on the Mac app, and adds new keyboard features for iOS users.
AI-generated summary ElevenLabs details voice AI infrastructure layers including speech recognition, turn-taking, and large language models for voice agents. ElevenLabs says its ElevenAgents uses Scribe v2 Realtime for speech transcription and supports Custom and Realtime modes.
AI-generated summary The author built a local voice-to-task system using Whisper to transcribe voice notes and Qwen running through llama.cpp to generate structured to-do items. Claude Code assisted with the development, allowing the entire pipeline to be packaged into a Python file with a graphical desktop user interface.
AI-generated summary Voxtral 3 features end-of-turn detection to help developers identify when a user has finished speaking. According to @MistralAI, the tool makes it easier to build responsive voice interactions.
AI-generated summary Forbes published an article by technology executive and software architect Alexandro Picolini discussing the engineering challenges of moving voice AI from a controlled demo environment into production. The author notes that real telephone operations face issues such as degraded audio quality, background noise, and latency that do not appear in simple demonstrations.
The top AI news in your email delivered each morning. Choose categories