In early June 2025, PlayAI open-sourced PlayDiffusion, a diffusion-based large language model (LLM) for audio speech editing, under the Apache 2.0 license. PlayDiffusion enables dynamic, fine-grained editing of speech audio without regenerating entire clips, preserving prosody, timing, and speaker identity. The model supports super-fine in-painting edits, zero-shot voice cloning, and generates audio in just 20-30 tokens, representing up to a 50-fold efficiency improvement over traditional autoregressive models. Zoom has expanded its AI Companion with over 45 new features, including improved real-time transcription, automated meeting summaries, action item detection, multilingual support, and deeper integration with business applications. These enhancements have contributed to an upward revision of Zoom's annual profit forecasts. AssemblyAI released a new real-time transcription model that offers greater speed and accuracy for speech-to-text applications. Speak AI has improved speaker identification and labeling for meetings, now supporting over 100 languages, with more than 200,000 users. Speak AI claims 95%+ transcription accuracy, 80%+ time savings, and a 4.9 G2 rating, and integrates with platforms such as Zoom, Microsoft Teams, Webex, and Google Meet. These developments demonstrate the rapid integration of advanced AI technologies into speech editing, transcription, and meeting productivity tools for professional and research use cases.