Tag
ElevenLabs 推出 Cue，一款强制对齐模型，可将音频中的每个字符精确到亚毫秒级时间线，准确度超过其测试的所有对齐器，覆盖 29 种语言，且一小时内音频的对齐耗时不到一秒。
NVIDIA released Nemotron 3.5 Transcribe, an open-weight promptable model for multilingual transcription, speaker diarization, translation, and audio Q&A with top leaderboard results, alongside the Granary v2 data pipeline and NeMo fine-tuning recipes for domain adaptation.
ETC-Bench is the first comprehensive multilingual benchmark for evaluation-side transcript canonicalization, covering 181 phenomenon types across 12 languages, along with a new SENA metric and a taxonomy-guided agent method (TACS) that outperforms baseline systems.
Mistral 发布 Voxtral 3 语音模型家族与语音代理 SDK，支持 50+ 语言的实时转录、情感自适应语音合成及多步工具调用，可通过欧盟托管 API 或自部署使用以保障数据主权。
SpeechFast 是一款 Mac 本地语音转文字工具，按住 fn 键即可在任意应用中输入，所有音频与文本均在本地处理、完全离线，并支持 Apple Intelligence 清理口头语及自定义词表。
IBM 发布 Granite Switch 4.1 30B/8B，这是一个模块化 LLM，将基座模型、多个 aLoRA 适配器（决策、guardian）和语音编码器组合在一起，通过单个端点即可在聊天、通用/专业决策和语音转文本之间切换，采用 Apache 2.0 许可。
Maybloc launches as a new state-of-the-art speech-to-text product for multi-speaker audio, claiming the lowest diarization error across several benchmarks, supporting 45 languages and up to 25 speakers, with add-ons like subtitles, PII redaction, summaries, and translation.
实践者分享构建电话语音AI代理的核心瓶颈并非模型能力，而是轮次切换时机：单靠VAD静音检测会导致打断或迟钝，改用STT置信度与部分转录文本的语义线索来判断发言是否结束效果更好。
Lipwise 是一款开源的跨平台智能听写应用，本地运行语音识别并在输出前用 LLM（如 Qwen3.5）自动修正、标点和改写文本，同时支持选中文本的指令式变换。应用采用 Rust 和 Tauri 构建，MIT 许可，无账户、无遥测，安装包约 5 MB。
Google's EmbeddingGemma 2 embedding model has been repurposed into a speech-to-text tool that transcribes an hour of audio in about 13 seconds entirely on-device on a base MacBook Air, using a small trained reader on top of the frozen audio encoder.
Show HN: BetterWispr 是一款免费、开源的 macOS 本地语音听写应用，支持在任意应用中按住快捷键说话，内置 Whisper、Parakeet 等九个本地模型，语音数据不离本机，并提供应用级语气风格和会议笔记功能。
一个基于 Qwen3.5 4B 的本地语音助手项目，支持技能与工具调用，无需独立 GPU，集成 Voxtral 本地语音识别与 Pocket TTS 语音合成。
A practitioner reflection on why voice agents cut callers off, arguing the real culprit is often VAD settings misreading background noise rather than STT latency, and that tuning silence thresholds changes how human an agent feels.
OpenWhispr launches its iOS app, offering 100% open source, on-device AI speech-to-text transcription that runs locally without accounts, subscriptions, or usage limits.
NihonSub is an open-source real-time Japanese anime subtitle and translation engine combining Whisper-Large-v3 on Groq with DeepSeek/LLaMA-3 for contextual bilingual subtitles, featuring VAD-based audio chunking, honorific-preserving LLM translation, and a synchronized dual-subtitle cinema player.
作者分享为电话语音机器人调优对话轮替（turn-taking）的经验：单靠 VAD 静音阈值不够，需结合 STT 部分文本判断句子是否完整，再配合 TTS 延迟一起调试，才能让机器人响应更自然。
Cactus Compute released Whistle, a 16.9 MB speech-to-text model running on CPU within the same engine as Needle, supporting seven languages and claiming mostly better accuracy than Whisper base at 9x smaller size and 6x speed.
Angebotsmeister 是一款面向德国工匠行业的语音报价工具：工人在现场口述内容即可自动生成包含明细和价格来源的报价单，客户可在线接受 PDF 报价，一键生成发票，免费起步。
Cactus-Compute 发布 Whistle，一个仅 16.9 MB 的端侧语音转文字模型，支持转录、词级时间戳和语音嵌入，可在手机、可穿戴设备、机器人及微控制器上无 GPU 运行，并在多项基准上与 Whisper 和 Moonshine 竞争。
Krisp released an open benchmark and dataset showing that voice isolation reduces word error rates in speech-to-text models by 73%, with significant improvements across workplace and call-center recordings.