Voxtral 3(mistral.ai)
747 points by ilreb | 200 comments
Mistral AI announcement page for Voxtral 3, its speech and audio AI model.
#speech-recognition on Hacker News
18 stories tagged speech-recognition, newest first.
Show HN: How random are you?(lab.specr.net)
205 points by vunderba | 108 comments
Interactive web game where you say or type numbers as randomly as you can while the page predicts each next number before you reveal it.
Whistle: Speech to Text in 16.9 MB(cactuscompute.com)
913 points by gmays | 176 comments
Cactus blog post announcing Whistle, a 16.9 MB open speech-to-text model for seven languages that runs on CPU alongside their Needle model.
Speech Recognition and TTS in less than 500kb(github.com)
544 points by petewarden | 79 comments
GitHub directory for Moonshine's micro build, offering speech recognition and neural text-to-speech small enough to run on microcontrollers like the RP2350.
Apple's new SpeechAnalyzer API, benchmarked against Whisper and its predecessor(get-inscribe.com)
529 points by get-inscribe | 211 comments
Blog post from Inscribe benchmarking Apple's new SpeechAnalyzer API against its predecessor and OpenAI Whisper on 5,559 test utterances, with transcripts released.
VibeVoice: Open-source frontier voice AI(github.com)
346 points by tosh | 169 comments
GitHub repository for Microsoft's VibeVoice, an open-source frontier voice AI project covering TTS and ASR models.
Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser(github.com)
374 points by Curiositry | 53 comments
GitHub repository for a Rust implementation of Mistral's Voxtral Mini 4B Realtime speech model that can run in the browser, with Q4/Q8 quantization.
Voxtral Transcribe 2(mistral.ai)
961 points by meetpateltech | 232 comments
Mistral AI announcement page for Voxtral Transcribe 2, its speech-to-text transcription model.
Show HN: I trained a 9M speech model to fix my Mandarin tones(simedw.com)
426 points by simedw | 128 comments
Blog post describing a 9M-parameter Conformer-CTC model trained on ~300 hours of speech that grades Mandarin pronunciation and tones in the browser.
FFmpeg 8.0 adds Whisper support(code.ffmpeg.org)
954 points by rilawa | 307 comments
FFmpeg commit adding a libavfilter audio filter that runs speech transcription with the Whisper model, including documentation and examples.
Complete silence is always hallucinated as "ترجمة نانسي قنقر" in Arabic(github.com)
504 points by edent | 283 comments
GitHub discussion reporting that OpenAI Whisper transcribes complete silence in Arabic as "Translation by Nancy Qunqar", with ffmpeg steps to reproduce.
Show HN: Real-time AI Voice Chat at ~500ms Latency(github.com)
480 points by koljab | 202 comments
GitHub repository for RealtimeVoiceChat, an open-source project for natural spoken conversation with an AI at low latency.
Launch HN: Aqua Voice (YC W24) – Voice-driven text editor
687 points by the_king | 230 comments
Launch HN post introducing Aqua Voice (YC W24), a voice-native document editor combining dictation with natural-language editing commands via LLMs.
Talk-Llama(github.com)
428 points by plurby | 126 comments
GitHub directory for talk-llama, a whisper.cpp example that lets you talk to a LLaMA model locally using speech recognition and text-to-speech.
Meta AI announces Massive Multilingual Speech code, models for 1000+ languages(github.com)
615 points by crakenzak | 193 comments
GitHub directory in Meta's fairseq repository containing code and models for Massive Multilingual Speech (MMS), covering ASR, TTS, and language ID for 1000+ languages.
Show HN: Willow – Open-source privacy-focused voice assistant hardware(github.com)
492 points by kkielhofner | 121 comments
GitHub repository for Willow, an open-source, self-hosted voice assistant firmware for Espressif ESP BOX hardware that integrates with Home Assistant.
Universal Speech Model(sites.research.google)
487 points by yamrzou | 181 comments
Google Research project page for the Universal Speech Model, a 2B-parameter ASR model family trained on 12 million hours of speech covering 300+ languages.
Launch HN: Vocode (YC W23) – Library for voice conversation with LLMs
322 points by KianHooshmand | 111 comments
Launch HN post for Vocode, a YC W23 open source Python library for building real-time voice conversation apps on top of LLMs.