Microsoft's MIT licensed VibeVoice speech-to-text model (think Whisper with speaker diarization) is really good - my notes on running the 5.71GB 4bit MLX conversion on an M5 MacBook, using about 60GB of RAM at peak and transcribing 1hr of audio in ~9 mins simonwillison.net/2026/Apr/27/...
microsoft/VibeVoice
VibeVoice is Microsoft's Whisper-style audio model for speech-to-text, MIT licensed and with speaker diarization built into the model. Microsoft released it on January 21st, 2026 but I hadn't tried it...
simonwillison.net
April 27, 2026 at 11:49 PM
Microsoft's MIT licensed VibeVoice speech-to-text model (think Whisper with speaker diarization) is really good - my notes on running the 5.71GB 4bit MLX conversion on an M5 MacBook, using about 60GB of RAM at peak and transcribing 1hr of audio in ~9 mins simonwillison.net/2026/Apr/27/...
Microsoft probably want to get the VibeVoice people to fix their Github security disclosure process
It says to contact MSRC, MSRC say they don't cover VibeVoice and close the reports.
It says to contact MSRC, MSRC say they don't cover VibeVoice and close the reports.
April 30, 2026 at 11:30 PM
Microsoft probably want to get the VibeVoice people to fix their Github security disclosure process
It says to contact MSRC, MSRC say they don't cover VibeVoice and close the reports.
It says to contact MSRC, MSRC say they don't cover VibeVoice and close the reports.
Since publishing my blog, Yaoyao Chang, who authored Vibing, has removed references to it from Microsoft’s VibeVoice repo - marking the change as “removing outdated links”. github.com/microsoft/Vi...
April 23, 2026 at 11:21 PM
Since publishing my blog, Yaoyao Chang, who authored Vibing, has removed references to it from Microsoft’s VibeVoice repo - marking the change as “removing outdated links”. github.com/microsoft/Vi...
Wow VibeVoice ASR 7B (!!) is absolute trash. Hallucinations, token loops, zero resilience to background noise. Another L from Microsoft who are quick to pat themselves on the back for essentially nothing of value. Qwen3 ASR 1.7B runs circles around it still, or Granite 4.1. Frontier my ass.
September 11, 2026 at 8:38 AM
Wow VibeVoice ASR 7B (!!) is absolute trash. Hallucinations, token loops, zero resilience to background noise. Another L from Microsoft who are quick to pat themselves on the back for essentially nothing of value. Qwen3 ASR 1.7B runs circles around it still, or Granite 4.1. Frontier my ass.
The package family name for Vibing.exe is YaoyaoChang.Vibing_ssp53fcyfr9ha - the same MSFT employee who digitally signed it, and posted GitHub messages calling it a “good product” he apparently knew nothing about on the MS VibeVoice GitHub repo. 😬😅
April 23, 2026 at 6:01 PM
The package family name for Vibing.exe is YaoyaoChang.Vibing_ssp53fcyfr9ha - the same MSFT employee who digitally signed it, and posted GitHub messages calling it a “good product” he apparently knew nothing about on the MS VibeVoice GitHub repo. 😬😅
🚀 Skyrocketing！ 🚀 (200+ new stars)
📦 microsoft / VibeVoice
⭐ 24,384 (+320)
🗒 Python
Open-Source Frontier Voice AI
📦 microsoft / VibeVoice
⭐ 24,384 (+320)
🗒 Python
Open-Source Frontier Voice AI
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
Open-Source Frontier Voice AI. Contribute to microsoft/VibeVoice development by creating an account on GitHub.
github.com
March 27, 2026 at 7:06 PM
Here's a uv one-liner that downloads and runs the MLX model against a local mp3 file
uv run --with mlx-audio python -m mlx_audio.stt.generate \
--model mlx-community/VibeVoice-ASR-4bit \
--audio lenny.mp3 --output-path lenny \
--format json --verbose --max-tokens 32768
uv run --with mlx-audio python -m mlx_audio.stt.generate \
--model mlx-community/VibeVoice-ASR-4bit \
--audio lenny.mp3 --output-path lenny \
--format json --verbose --max-tokens 32768
April 27, 2026 at 11:53 PM
Here's a uv one-liner that downloads and runs the MLX model against a local mp3 file
uv run --with mlx-audio python -m mlx_audio.stt.generate \
--model mlx-community/VibeVoice-ASR-4bit \
--audio lenny.mp3 --output-path lenny \
--format json --verbose --max-tokens 32768
uv run --with mlx-audio python -m mlx_audio.stt.generate \
--model mlx-community/VibeVoice-ASR-4bit \
--audio lenny.mp3 --output-path lenny \
--format json --verbose --max-tokens 32768
I'm trying out VibeVoice (Microsoft ASR for speech-to-text) for a project I want to work on. Testing it with some @deadblossomjesse.bsky.social clips, this is how it interprets their laugh.
May 30, 2026 at 3:29 PM
I'm trying out VibeVoice (Microsoft ASR for speech-to-text) for a project I want to work on. Testing it with some @deadblossomjesse.bsky.social clips, this is how it interprets their laugh.
Deploy VibeVoice 1.5B locally with audio.cpp! Using a native C++/ggml runtime on NVIDIA GPUs (like the RTX 5090) delivers ~2.86x speedup over Python, reaching up to 4.08x faster-than-real-time speeds. #VibeVoice #CUDA #GPU #Cplusplus #MachineLearning https://ai.dappcrypto.org/r/n
July 1, 2026 at 2:04 PM
Deploy VibeVoice 1.5B locally with audio.cpp! Using a native C++/ggml runtime on NVIDIA GPUs (like the RTX 5090) delivers ~2.86x speedup over Python, reaching up to 4.08x faster-than-real-time speeds. #VibeVoice #CUDA #GPU #Cplusplus #MachineLearning https://ai.dappcrypto.org/r/n
you can make the VibeVoice German voice models lecture you in English about things you couldn't possibly know
it's stupid but it amuses me
it's stupid but it amuses me
January 4, 2026 at 2:00 AM
you can make the VibeVoice German voice models lecture you in English about things you couldn't possibly know
it's stupid but it amuses me
it's stupid but it amuses me
Microsoft VibeVoice: Open-Source Frontier Voice AI
https://github.com/microsoft/VibeVoice
https://news.ycombinator.com/item?id=47933236
https://github.com/microsoft/VibeVoice
https://news.ycombinator.com/item?id=47933236
April 28, 2026 at 7:20 PM
Microsoft VibeVoice: Open-Source Frontier Voice AI
https://github.com/microsoft/VibeVoice
https://news.ycombinator.com/item?id=47933236
https://github.com/microsoft/VibeVoice
https://news.ycombinator.com/item?id=47933236
🔥 GitHub Trending — Wednesday, April 29, 2026
#1 skills · #2 free-claude-code · #3 GitNexus · #4 VibeVoice · #5 GhostTrack
#GitHub #OpenSource #Dev
#1 skills · #2 free-claude-code · #3 GitNexus · #4 VibeVoice · #5 GhostTrack
#GitHub #OpenSource #Dev
April 29, 2026 at 6:00 AM
🔥 GitHub Trending — Wednesday, April 29, 2026
#1 skills · #2 free-claude-code · #3 GitNexus · #4 VibeVoice · #5 GhostTrack
#GitHub #OpenSource #Dev
#1 skills · #2 free-claude-code · #3 GitNexus · #4 VibeVoice · #5 GhostTrack
#GitHub #OpenSource #Dev
If you thought there were "too many podcasts" already, get ready for a massive uptick in "slopcasts"
VibeVoice is designed for generating expressive, long-form, multi-speaker conversational audio, such as podcasts, from text.
VibeVoice is designed for generating expressive, long-form, multi-speaker conversational audio, such as podcasts, from text.
GitHub - microsoft/VibeVoice: Frontier Open-Source Text-to-Speech
Frontier Open-Source Text-to-Speech. Contribute to microsoft/VibeVoice development by creating an account on GitHub.
github.com
August 26, 2025 at 1:00 PM
If you thought there were "too many podcasts" already, get ready for a massive uptick in "slopcasts"
VibeVoice is designed for generating expressive, long-form, multi-speaker conversational audio, such as podcasts, from text.
VibeVoice is designed for generating expressive, long-form, multi-speaker conversational audio, such as podcasts, from text.
#Microsoft revolutionizes #audio with #VibeVoice #YuriOntibon #AIVoice #Innovation #Software #AI #SatyaNadella
February 5, 2026 at 2:21 AM
#Microsoft revolutionizes #audio with #VibeVoice #YuriOntibon #AIVoice #Innovation #Software #AI #SatyaNadella
Запускайте VibeVoice 1.5B локально на RTX 5090! Использование audio.cpp и ggml дает ускорение в ~2.86x по сравнению с Python, обеспечивая генерацию быстрее реального времени без лишних зависимостей. #VibeVoice #NVIDIA #CUDA #AI #GPU https://ai.dappcrypto.org/r/n
July 1, 2026 at 2:11 PM
Запускайте VibeVoice 1.5B локально на RTX 5090! Использование audio.cpp и ggml дает ускорение в ~2.86x по сравнению с Python, обеспечивая генерацию быстрее реального времени без лишних зависимостей. #VibeVoice #NVIDIA #CUDA #AI #GPU https://ai.dappcrypto.org/r/n
ComfyUI integration for Microsoft VibeVoice TTS
December 3, 2025 at 8:33 AM
ComfyUI integration for Microsoft VibeVoice TTS
VibeVoice is a family of #opensource frontier #voiceAI models that includes both Text-to-Speech (TTS) and Automatic Speech Recognition (ASR) models.
github.com/microsoft/Vi...
#AI #github #microsoft
github.com/microsoft/Vi...
#AI #github #microsoft
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
Open-Source Frontier Voice AI. Contribute to microsoft/VibeVoice development by creating an account on GitHub.
github.com
April 28, 2026 at 2:43 PM
VibeVoice is a family of #opensource frontier #voiceAI models that includes both Text-to-Speech (TTS) and Automatic Speech Recognition (ASR) models.
github.com/microsoft/Vi...
#AI #github #microsoft
github.com/microsoft/Vi...
#AI #github #microsoft
今日のGitHubトレンド
microsoft/VibeVoice
VibeVoiceは、Microsoftによるオープンソースの音声AI研究フレームワークです。
テキスト読み上げ（TTS）と音声認識（ASR）モデルを公開し、長時間の音声処理や多言語対応、リアルタイム性を実現しています。
音声合成コミュニティの協力を促進し、AIの責任ある利用を目指しています。
microsoft/VibeVoice
VibeVoiceは、Microsoftによるオープンソースの音声AI研究フレームワークです。
テキスト読み上げ（TTS）と音声認識（ASR）モデルを公開し、長時間の音声処理や多言語対応、リアルタイム性を実現しています。
音声合成コミュニティの協力を促進し、AIの責任ある利用を目指しています。
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
Open-Source Frontier Voice AI. Contribute to microsoft/VibeVoice development by creating an account on GitHub.
github.com
January 25, 2026 at 11:18 AM
今日のGitHubトレンド
microsoft/VibeVoice
VibeVoiceは、Microsoftによるオープンソースの音声AI研究フレームワークです。
テキスト読み上げ（TTS）と音声認識（ASR）モデルを公開し、長時間の音声処理や多言語対応、リアルタイム性を実現しています。
音声合成コミュニティの協力を促進し、AIの責任ある利用を目指しています。
microsoft/VibeVoice
VibeVoiceは、Microsoftによるオープンソースの音声AI研究フレームワークです。
テキスト読み上げ（TTS）と音声認識（ASR）モデルを公開し、長時間の音声処理や多言語対応、リアルタイム性を実現しています。
音声合成コミュニティの協力を促進し、AIの責任ある利用を目指しています。
I've been experiment with using voice-to-text engines to read my own stories to me. It's helpful as hell in identifying typos and awkward phrases. As an experiment, I installed VibeVoice from Microsoft. It was okay with the default voice, but I decided to try a further experiment.
November 3, 2025 at 7:50 PM
I've been experiment with using voice-to-text engines to read my own stories to me. It's helpful as hell in identifying typos and awkward phrases. As an experiment, I installed VibeVoice from Microsoft. It was okay with the default voice, but I decided to try a further experiment.
Not JUST music, but a lowfi approximation of the quasi-martial adventure music used at the beginning of every Honor Harrington audiobook. The VibeVoice-70b model apparently "knows" that whenever Allison Johnson reads something to you, it starts with a few seconds of underlying music.
November 3, 2025 at 7:50 PM
Not JUST music, but a lowfi approximation of the quasi-martial adventure music used at the beginning of every Honor Harrington audiobook. The VibeVoice-70b model apparently "knows" that whenever Allison Johnson reads something to you, it starts with a few seconds of underlying music.
v1.4.30 is out
It's time for another update, here's the highlights: Hotfix to VibeVoice: Errors on startup show properly and on languageswhere decimal separator is ',' instead of '.', it now works…sigh… 😀 Fix issue of not all audio rendering when playback markers were used VibeVoice upgraded to…
It's time for another update, here's the highlights: Hotfix to VibeVoice: Errors on startup show properly and on languageswhere decimal separator is ',' instead of '.', it now works…sigh… 😀 Fix issue of not all audio rendering when playback markers were used VibeVoice upgraded to…
v1.4.30 is out
It's time for another update, here's the highlights: Hotfix to VibeVoice: Errors on startup show properly and on languageswhere decimal separator is ',' instead of '.', it now works…sigh… 😀 Fix issue of not all audio rendering when playback markers were used VibeVoice upgraded to Community edition: 1.5B & 7B models supportedVoice cloning also as well as cfg scale & step amount…
lyricvideo.studio
January 20, 2026 at 2:42 PM
v1.4.30 is out
It's time for another update, here's the highlights: Hotfix to VibeVoice: Errors on startup show properly and on languageswhere decimal separator is ',' instead of '.', it now works…sigh… 😀 Fix issue of not all audio rendering when playback markers were used VibeVoice upgraded to…
It's time for another update, here's the highlights: Hotfix to VibeVoice: Errors on startup show properly and on languageswhere decimal separator is ',' instead of '.', it now works…sigh… 😀 Fix issue of not all audio rendering when playback markers were used VibeVoice upgraded to…
Microsoft's MIT licensed VibeVoice speech-to-text model (think Whisper with speaker diarization) is really good - my notes on running the 5.71GB 4bit MLX conversion on an M5 MacBook, using about 60GB of RAM at peak and transcribing 1hr of audio in ~9 mins […]
Original post on fedi.simonwillison.net
fedi.simonwillison.net
April 27, 2026 at 11:48 PM
Microsoft's MIT licensed VibeVoice speech-to-text model (think Whisper with speaker diarization) is really good - my notes on running the 5.71GB 4bit MLX conversion on an M5 MacBook, using about 60GB of RAM at peak and transcribing 1hr of audio in ~9 mins […]
<a href="https://github.com/microsoft/VibeVoice" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">github.com/microsoft/VibeVoice
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
github.com
December 14, 2025 at 12:45 PM
<a href="https://github.com/microsoft/VibeVoice" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">github.com/microsoft/VibeVoice
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
GitHub - microsoft/VibeVoice: Open-Source Frontier Voice AI
マイクロソフト、音声合成モデル「VibeVoice」をオープンソースで公開、最大4名・90分までの自然な会話を生成
https://codezine.jp/news/detail/22175
https://codezine.jp/news/detail/22175
マイクロソフト、音声合成モデル「VibeVoice」をオープンソースで公開、最大4名・90分までの自然な会話を生成
マイクロソフトは、オープンソースのテキスト音声合成（TTS）モデル「VibeVoice」を2025年8月26日（現地時間）に公開した。
codezine.jp
August 30, 2025 at 2:48 AM
マイクロソフト、音声合成モデル「VibeVoice」をオープンソースで公開、最大4名・90分までの自然な会話を生成
https://codezine.jp/news/detail/22175
https://codezine.jp/news/detail/22175