{"slug": "sherpa-onnx", "name": "sherpa-onnx", "category": "ai", "category_name": "AI tools and apps", "description": "", "recommenders": 5, "commenters": 6, "warners": 0, "score": 4.623, "rising": -1.64, "first": "2025-11-12", "latest": "2026-10-29", "why": "So I switched over to Sherpa [3] trained on gigaspeech. It\u2019s significantly more accurate", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 5, "neutral": 1, "compare": 0, "warn": 0}, "spark": [1, 0, 0, 0, 2, 0, 1, 1, 0, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "So I switched over to Sherpa [3] trained on gigaspeech. It\u2019s significantly more accurate", "kind": "", "icon": true, "domain": null, "github_repo": "github.com/k2-fsa/sherpa-onnx", "official_url": "https://github.com/k2-fsa/sherpa-onnx", "pricing": "", "aliases": ["sherpa-onnx", "Sherpa-ONNX"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 773813840, "item_id": 48363365, "url": "https://news.ycombinator.com/item?id=48363365", "by": "denysvitali", "date": "2026-06-01", "version": null, "stance": "recommend", "quote": "https://k2-fsa.github.io/sherpa/onnx/index.html", "excerpt": "https://k2-fsa.github.io/sherpa/onnx/index.html", "context": "", "use_case": "Speech-to-text", "switched_from": "", "story": "GrapheneOS Speech Services version 2 released"}, {"id": 767988049, "item_id": 47999253, "url": "https://news.ycombinator.com/item?id=47999253", "by": "vunderba", "date": "2026-05-03", "version": null, "stance": "recommend", "quote": "So I switched over to Sherpa [3] trained on gigaspeech. It\u2019s significantly more accurate", "excerpt": "So I use a VAD onnx (Silero [1]) to automatically detect when someone is talking, and then it sends the audio into one of the voice recognition libraries. I originally tried to get away with just Whisper Tiny in the chess game [2], but it performs worse on the kinds of short phrases (knight E4, c takes d5, etc) used to dictate chess notation. Even with hotword-based phrasing and corrections, I found its accuracy on b", "context": "", "use_case": "voice dictation for chess game", "switched_from": "Whisper Tiny", "story": "Show HN: Apple's Sharp Running in the Browser via ONNX Runtime Web"}, {"id": 757296627, "item_id": 47331039, "url": "https://news.ycombinator.com/item?id=47331039", "by": "sanchitmonga22", "date": "2026-03-10", "version": null, "stance": "recommend", "quote": "it might be the best starting point for what your after", "excerpt": "Understood, you want dictation, not a chatbot. That's a valid and different use case. RCLI is Apple Silicon only today because MetalRT is built on Metal. For Linux, the closest thing to what you're describing would be building a virtual input device on top of Whisper or Parakeet (which RCLI supports as STT backends). Parakeet TDT 0.6B has ~1.9% WER, that's very close to production dictation quality. The missing piece", "context": "", "use_case": "streaming speech-to-text", "switched_from": "", "story": "Show HN: RunAnwhere \u2013 Faster AI Inference on Apple Silicon"}, {"id": 757248037, "item_id": 47328002, "url": "https://news.ycombinator.com/item?id=47328002", "by": "pzo", "date": "2026-03-10", "version": null, "stance": "recommend", "quote": "Other popular repos for such tasks I would recommend:", "excerpt": "FWIW this RCLI is only MIT license but their engine MetalRT is commercial. Not sure the license of their models I guess also not MIT. So IMHO this repo is misleading. Not sure why they decided to reinvent the wheel and write yet another ML engine (MetalRT) which is proprietary. I would most likely bet on CoreML since it have support for ANE (apple NPU) or MLX. Other popular repos for such tasks I would recommend: htt", "context": "", "use_case": "speech ML", "switched_from": "", "story": "Show HN: RunAnwhere \u2013 Faster AI Inference on Apple Silicon"}, {"id": 734498673, "item_id": 45906167, "url": "https://news.ycombinator.com/item?id=45906167", "by": "lostmsu", "date": "2025-11-12", "version": null, "stance": "recommend", "quote": "Kokoro is available as a system TTS for Android via OSS project called", "excerpt": "Kokoro is available as a system TTS for Android via OSS project called \"sherpa\": https://k2-fsa.github.io/sherpa/onnx/android/index.html I believe its power usage is negligible in comparison to, for example, screen or maybe even Bluetooth audio.", "context": "", "use_case": "", "switched_from": "", "story": "TTS still sucks"}]}], "caveats": [], "who": [], "tags": [], "use_cases": ["speech-to-text", "voice dictation for chess game", "streaming speech-to-text", "speech ml"], "compared": [{"slug": "kokoro", "name": "Kokoro", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "nvidia-parakeet", "name": "NVIDIA Parakeet", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [{"id": 761652672, "item_id": 47603292, "url": "https://news.ycombinator.com/item?id=47603292", "by": "Edison2ST", "date": "2026-04-01", "version": null, "stance": "recommend", "quote": "contributed to Sherpa-ONNX (11.2k stars)", "excerpt": "Location: Barcelona, Spain Remote: Yes Willing to relocate: Yes Technologies: ESP32 (ESP-IDF), C/C++, Rust, Python/Django, AI Optimization (LiteRT, ONNX), Reverse Engineering (Ghidra, gdb). R\u00e9sum\u00e9/CV: https://drive.proton.me/urls/YNHA7NPJ3M#Q4wLbDAP0LQ3 Email: edison2st (at) proton (dot) me Edison dos Santos | Barcelona, Spain (EU/PT citizen) | Remote or Relocation | Firmware \u00b7 Embedded Systems \u00b7 On-Device ML Built a", "context": "Embedded/firmware engineer seeking work", "use_case": "On-device inference", "switched_from": "", "story": "Ask HN: Who wants to be hired? (April 2026)"}], "threads": [{"url": "https://news.ycombinator.com/item?id=47326101", "title": "Show HN: RunAnwhere \u2013 Faster AI Inference on Apple Silicon", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=50287286", "title": "Show HN: Audio-chat-box \u2013 offline speech in and out for any program, over stdio", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48360871", "title": "GrapheneOS Speech Services version 2 released", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47995037", "title": "Show HN: Apple's Sharp Running in the Browser via ONNX Runtime Web", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47601858", "title": "Ask HN: Who wants to be hired? (April 2026)", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=45881279", "title": "TTS still sucks", "mentions": 1}], "sentiment_by_month": {"2025-11": {"recommend": 1}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {"recommend": 2}, "2026-04": {}, "2026-05": {"recommend": 1}, "2026-06": {"recommend": 1}, "2026-07": {}, "2026-08": {}, "2026-09": {}, "2026-10": {"neutral": 1}}, "versions": [], "version_unnamed": 0, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 804596627, "item_id": 50287289, "url": "https://news.ycombinator.com/item?id=50287289", "by": "cjus", "date": "2026-10-29", "version": null, "stance": "neutral", "quote": "It all runs locally through sherpa-onnx.", "excerpt": "I built audio-chat-box to give a program a voice without sending audio anywhere. It reads text aloud with Kokoro, transcribes with NVIDIA's Parakeet TDT 0.6B v2, and can listen hands-free: your program streams microphone samples, Silero VAD finds each thing said, and Parakeet transcribes it when the speaker pauses. It all runs locally through sherpa-onnx. To try it without building anything, there's a small browser d", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: Audio-chat-box \u2013 offline speech in and out for any program, over stdio"}, {"id": 773813840, "item_id": 48363365, "url": "https://news.ycombinator.com/item?id=48363365", "by": "denysvitali", "date": "2026-06-01", "version": null, "stance": "recommend", "quote": "https://k2-fsa.github.io/sherpa/onnx/index.html", "excerpt": "https://k2-fsa.github.io/sherpa/onnx/index.html", "context": "", "use_case": "Speech-to-text", "switched_from": "", "story": "GrapheneOS Speech Services version 2 released"}, {"id": 767988049, "item_id": 47999253, "url": "https://news.ycombinator.com/item?id=47999253", "by": "vunderba", "date": "2026-05-03", "version": null, "stance": "recommend", "quote": "So I switched over to Sherpa [3] trained on gigaspeech. It\u2019s significantly more accurate", "excerpt": "So I use a VAD onnx (Silero [1]) to automatically detect when someone is talking, and then it sends the audio into one of the voice recognition libraries. I originally tried to get away with just Whisper Tiny in the chess game [2], but it performs worse on the kinds of short phrases (knight E4, c takes d5, etc) used to dictate chess notation. Even with hotword-based phrasing and corrections, I found its accuracy on b", "context": "", "use_case": "voice dictation for chess game", "switched_from": "Whisper Tiny", "story": "Show HN: Apple's Sharp Running in the Browser via ONNX Runtime Web"}, {"id": 757296627, "item_id": 47331039, "url": "https://news.ycombinator.com/item?id=47331039", "by": "sanchitmonga22", "date": "2026-03-10", "version": null, "stance": "recommend", "quote": "it might be the best starting point for what your after", "excerpt": "Understood, you want dictation, not a chatbot. That's a valid and different use case. RCLI is Apple Silicon only today because MetalRT is built on Metal. For Linux, the closest thing to what you're describing would be building a virtual input device on top of Whisper or Parakeet (which RCLI supports as STT backends). Parakeet TDT 0.6B has ~1.9% WER, that's very close to production dictation quality. The missing piece", "context": "", "use_case": "streaming speech-to-text", "switched_from": "", "story": "Show HN: RunAnwhere \u2013 Faster AI Inference on Apple Silicon"}, {"id": 757248037, "item_id": 47328002, "url": "https://news.ycombinator.com/item?id=47328002", "by": "pzo", "date": "2026-03-10", "version": null, "stance": "recommend", "quote": "Other popular repos for such tasks I would recommend:", "excerpt": "FWIW this RCLI is only MIT license but their engine MetalRT is commercial. Not sure the license of their models I guess also not MIT. So IMHO this repo is misleading. Not sure why they decided to reinvent the wheel and write yet another ML engine (MetalRT) which is proprietary. I would most likely bet on CoreML since it have support for ANE (apple NPU) or MLX. Other popular repos for such tasks I would recommend: htt", "context": "", "use_case": "speech ML", "switched_from": "", "story": "Show HN: RunAnwhere \u2013 Faster AI Inference on Apple Silicon"}, {"id": 734498673, "item_id": 45906167, "url": "https://news.ycombinator.com/item?id=45906167", "by": "lostmsu", "date": "2025-11-12", "version": null, "stance": "recommend", "quote": "Kokoro is available as a system TTS for Android via OSS project called", "excerpt": "Kokoro is available as a system TTS for Android via OSS project called \"sherpa\": https://k2-fsa.github.io/sherpa/onnx/android/index.html I believe its power usage is negligible in comparison to, for example, screen or maybe even Bluetooth audio.", "context": "", "use_case": "", "switched_from": "", "story": "TTS still sucks"}]}