Tag

Handy is a free, open-source, extensible speech-to-text desktop application that works completely offline on Windows, macOS, and Linux. It leverages local AI models like Whisper and Parakeet V3 for privacy-focused transcription, allowing users to press a shortcut, speak, and paste text directly into any app without sending audio to the cloud. Key features include GPU acceleration, automatic language detection, and high extensibility for developers.

NotebookLM is an AI-powered research and note-taking tool developed by Google, utilizing Gemini models to analyze user-uploaded sources like documents, websites, and videos, generating summaries, insights, and audio overviews while prioritizing privacy.

WhisperLiveKit is an ultra-low-latency, self-hosted speech-to-text toolkit with speaker identification. Powered by leading simultaneous speech research like Simul-Whisper and WhisperStreaming, it enables intelligent buffering and incremental processing for real-time transcription, translation across 200 languages, and speaker diarization. Ideal for meeting notes, accessibility tools, and content creation.

VibeVoice is Microsoft's open-source frontier voice AI framework designed for generating expressive, long-form, multi-speaker conversational audio (e.g., podcasts) from text. It supports up to 90 minutes of speech with up to 4 distinct speakers. Key innovations include continuous speech tokenizers at 7.5 Hz frame rate and next-token diffusion using LLMs for context and high-fidelity acoustics. Recently released VibeVoice-Realtime-0.5B for real-time streaming TTS with ~300ms latency

This paper presents Deep Speech 2, an end-to-end deep learning system for automatic speech recognition that works across vastly different languages (English and Mandarin). It replaces traditional hand-engineered ASR pipelines with neural networks, achieving human-competitive transcription accuracy on standard datasets. The system uses HPC techniques for 7x speedup, enabling faster experimentation. Key innovations include Batch Normalization for RNNs, curriculum learning (SortaGrad), and GPU deployment optimization (Batch Dispatch). The approach demonstrates that end-to-end learning can handle diverse speech conditions including noise, accents, and different languages, representing a significant step toward universal speech recognition systems.