A recent write-up from researchers at HumeAI found that automatic speech recognition (ASR) models exhibiting benchmark-optimized behavior reproduce erroneous reference transcript text, indicating they…
Automatic subtitling powered by AI has become viable for production companies in 2026, with OpenAI Whisper v4 achieving over 97% accuracy in clean audio conditions for major languages and AssemblyAI o…
Speko, a San Francisco-based voice AI routing layer founded by Beknazar Abdikamalov in 2026, selects speech, language, and synthesis models using language-specific benchmarks, offering a /v1 API that …
A developer outlines a method for EU startups to select a speech-to-text API for supplier invoice transcription, emphasizing cost per accepted invoice over per-minute pricing. The approach involves bu…
AssemblyAI has published a comparison of OpenAI Realtime API alternatives for 2026, highlighting cost unpredictability, transcription accuracy issues, and barge-in quality as key pain points. The comp…
A developer at AssemblyAI argues that the best voice agent API is determined by speech-to-text accuracy, not flashy features, because errors in transcription propagate through the entire pipeline. The…
LiveKit Agents, the open-source framework from LiveKit that powers ChatGPT's voice mode and counts xAI, Salesforce, and 911 dispatch operators among its customers, is trending on GitHub as developers …
Two macOS apps, Vox Dictum and Bonfiyah, offer AI-powered transcription and summarization features, with Vox Dictum emphasizing on-device privacy and Bonfiyah providing conversation analysis and actio…
Granola, an AI note-taking app, documents a SOC 2 Type II certification, encryption in transit and at rest, US AWS hosting, and an organization-wide model-training opt-out, but states it is not HIPAA …
AssemblyAI's 2026 guide to building voice-controlled AI agents identifies orchestration—latency, turn-taking, tool calls, and interruption handling—as the core engineering challenge, not the prompt or…
Willow, a voice dictation startup, announced a new platform on July 23rd, 2026, that measures success by a "zero edit rate" — the proportion of transcriptions that require no user edits before sending…
AssemblyAI's Universal-3.5 Pro Realtime streaming speech-to-text model achieves a 6.99% pooled word error rate on Pipecat's open STT benchmark, outperforming competitors like Deepgram Flux (15.58%), E…
A developer at AssemblyAI argues that healthcare AI teams face a strategic choice between buying finished ambient clinical documentation products like Nuance DAX Copilot or Abridge and building their …
A developer outlines three architectures for building voice agents: cascading (STT→LLM→TTS), speech-to-speech (multimodal model), and unified API. The post emphasizes that accuracy depends on the spee…
XAI launched the Grok Voice Agent Builder in beta on July 1, featuring a single speech-to-speech model that replaces the traditional three-API voice stack. The platform offers sub-700ms latency and co…
A developer released e3d-pod2vid, an open-source AI pipeline that converts diarized podcast audio into YouTube-ready videos with semantically matched B-roll, burned-in subtitles, and optional TTS voic…
SouthBridgeAI released offmute-v2, an open-source multi-step pipeline for timestamp-correct, diarized transcripts with identified speakers, after GLM-5.2 outperformed Opus 4.8 on an AI-resistant backe…
LiveKit released eot-bench, an open benchmark suite for end-of-turn detection in voice AI, along with the first open dataset of real human-to-agent conversations in 14 languages. The benchmark evaluat…
Whissle Research Team introduced a meta-aware automatic speech recognition (ASR) model, META-1, that outputs transcription and metadata tokens (emotion, intent, demographics) in a single forward pass …
A developer benchmarked five voice AI stacks using the same one-minute conversation and found that only two—OpenAI's Realtime API and LiveKit Agents with Gemini 2.0 Flash Live—consistently stayed unde…