Hacker News: watsonmusichttps://news.ycombinator.com/user?id=watsonmusicHacker News RSShttps://hnrss.org/hnrss v2.1.1Wed, 29 Apr 2026 18:48:45 +0000<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>it's not oss

]]>Wed, 03 Sep 2025 15:04:47 +0000https://news.ycombinator.com/item?id=45116659watsonmusichttps://news.ycombinator.com/item?id=45116659https://news.ycombinator.com/item?id=45116659<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>bonus usage

]]>Wed, 03 Sep 2025 15:04:27 +0000https://news.ycombinator.com/item?id=45116655watsonmusichttps://news.ycombinator.com/item?id=45116655https://news.ycombinator.com/item?id=45116655<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>11labs is facing a real competitor

]]>Wed, 03 Sep 2025 14:58:36 +0000https://news.ycombinator.com/item?id=45116592watsonmusichttps://news.ycombinator.com/item?id=45116592https://news.ycombinator.com/item?id=45116592<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>genius

]]>Wed, 03 Sep 2025 14:57:34 +0000https://news.ycombinator.com/item?id=45116577watsonmusichttps://news.ycombinator.com/item?id=45116577https://news.ycombinator.com/item?id=45116577<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>this model is superb

]]>Wed, 03 Sep 2025 14:56:50 +0000https://news.ycombinator.com/item?id=45116568watsonmusichttps://news.ycombinator.com/item?id=45116568https://news.ycombinator.com/item?id=45116568<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>Microsoft is cool

]]>Wed, 03 Sep 2025 14:56:08 +0000https://news.ycombinator.com/item?id=45116560watsonmusichttps://news.ycombinator.com/item?id=45116560https://news.ycombinator.com/item?id=45116560<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>yes the best

]]>Wed, 03 Sep 2025 14:53:53 +0000https://news.ycombinator.com/item?id=45116531watsonmusichttps://news.ycombinator.com/item?id=45116531https://news.ycombinator.com/item?id=45116531<![CDATA[New comment by watsonmusic in "VibeVoice: A Frontier Open-Source Text-to-Speech Model"]]>one of the best models built by Microsoft

]]>Wed, 03 Sep 2025 14:51:54 +0000https://news.ycombinator.com/item?id=45116509watsonmusichttps://news.ycombinator.com/item?id=45116509https://news.ycombinator.com/item?id=45116509<![CDATA[New comment by watsonmusic in "Microsoft releases VibeVoice, generates 90-minute, 4-speaker audio"]]>https://github.com/microsoft/VibeVoice

]]>Tue, 26 Aug 2025 13:25:46 +0000https://news.ycombinator.com/item?id=45026227watsonmusichttps://news.ycombinator.com/item?id=45026227https://news.ycombinator.com/item?id=45026227<![CDATA[New comment by watsonmusic in "Microsoft releases VibeVoice, generates 90-minute, 4-speaker audio"]]>https://huggingface.co/microsoft/VibeVoice-1.5B

]]>Tue, 26 Aug 2025 13:25:35 +0000https://news.ycombinator.com/item?id=45026226watsonmusichttps://news.ycombinator.com/item?id=45026226https://news.ycombinator.com/item?id=45026226<![CDATA[New comment by watsonmusic in "Microsoft releases VibeVoice, generates 90-minute, 4-speaker audio"]]>VibeVoice is a novel framework designed for generating expressive, long-form, multi-speaker conversational audio, such as podcasts, from text. It addresses significant challenges in traditional Text-to-Speech (TTS) systems, particularly in scalability, speaker consistency, and natural turn-taking. A core innovation of VibeVoice is its use of continuous speech tokenizers (Acoustic and Semantic) operating at an ultra-low frame rate of 7.5 Hz. These tokenizers efficiently preserve audio fidelity while significantly boosting computational efficiency for processing long sequences. VibeVoice employs a next-token diffusion framework, leveraging a Large Language Model (LLM) to understand textual context and dialogue flow, and a diffusion head to generate high-fidelity acoustic details. The model can synthesize speech up to 90 minutes long with up to 4 distinct speakers, surpassing the typical 1-2 speaker limits of many prior models.

]]>Tue, 26 Aug 2025 13:24:42 +0000https://news.ycombinator.com/item?id=45026219watsonmusichttps://news.ycombinator.com/item?id=45026219https://news.ycombinator.com/item?id=45026219<![CDATA[Microsoft releases VibeVoice, generates 90-minute, 4-speaker audio]]>Article URL: https://microsoft.github.io/VibeVoice/

Comments URL: https://news.ycombinator.com/item?id=45026218

Points: 3

# Comments: 3

]]>Tue, 26 Aug 2025 13:24:42 +0000https://microsoft.github.io/VibeVoice/watsonmusichttps://news.ycombinator.com/item?id=45026218https://news.ycombinator.com/item?id=45026218<![CDATA[New comment by watsonmusic in "Reinforcement Pre-Training"]]>cannot wait seeing how it goes beyond the current llm training pipeline

]]>Tue, 10 Jun 2025 16:39:25 +0000https://news.ycombinator.com/item?id=44238819watsonmusichttps://news.ycombinator.com/item?id=44238819https://news.ycombinator.com/item?id=44238819<![CDATA[New comment by watsonmusic in "Reinforcement Pre-Training"]]>it could be adaptive. only high-value tokens were allocated with more compute

]]>Tue, 10 Jun 2025 16:37:41 +0000https://news.ycombinator.com/item?id=44238797watsonmusichttps://news.ycombinator.com/item?id=44238797https://news.ycombinator.com/item?id=44238797<![CDATA[New comment by watsonmusic in "Reinforcement Pre-Training"]]>A new scaling paradigm finally comes out!

]]>Tue, 10 Jun 2025 16:34:49 +0000https://news.ycombinator.com/item?id=44238774watsonmusichttps://news.ycombinator.com/item?id=44238774https://news.ycombinator.com/item?id=44238774<![CDATA[New comment by watsonmusic in "Reinforcement Pre-Training"]]>14b model performs comparably with 32b size. the improvement is huge

]]>Tue, 10 Jun 2025 16:33:04 +0000https://news.ycombinator.com/item?id=44238758watsonmusichttps://news.ycombinator.com/item?id=44238758https://news.ycombinator.com/item?id=44238758<![CDATA[New comment by watsonmusic in "Differential Transformer"]]>negative values can enhance the expressibility

]]>Tue, 08 Oct 2024 17:37:10 +0000https://news.ycombinator.com/item?id=41779779watsonmusichttps://news.ycombinator.com/item?id=41779779https://news.ycombinator.com/item?id=41779779<![CDATA[New comment by watsonmusic in "Differential Transformer"]]>not all hallucinations are creativity Imaginate that for a RAG application, the model is supposed to follow the given documents

]]>Tue, 08 Oct 2024 13:56:48 +0000https://news.ycombinator.com/item?id=41777437watsonmusichttps://news.ycombinator.com/item?id=41777437https://news.ycombinator.com/item?id=41777437<![CDATA[New comment by watsonmusic in "Differential Transformer"]]>the model is supposed to learn this

]]>Tue, 08 Oct 2024 13:05:02 +0000https://news.ycombinator.com/item?id=41776928watsonmusichttps://news.ycombinator.com/item?id=41776928https://news.ycombinator.com/item?id=41776928<![CDATA[New comment by watsonmusic in "Differential Transformer"]]>The modification is simple and beautiful. And the improvements are quite significant.

]]>Tue, 08 Oct 2024 13:02:48 +0000https://news.ycombinator.com/item?id=41776910watsonmusichttps://news.ycombinator.com/item?id=41776910https://news.ycombinator.com/item?id=41776910