Upload 9 files

realtime_model/text_examples/1p_abs.txt

@@ -0,0 +1,2 @@
+Generating long-form, multi-speaker conversational audio like podcasts poses significant challenges for traditional Text-to-Speech (TTS) systems, particularly in scalability, speaker consistency, and natural turn-taking. This report presents VibeVoice, a novel model designed to synthesize long-form speech with multiple speakers by employing the next-token diffusion framework, a unified method for modeling continuous data by autoregressively generating latent vectors via diffusion. 
+A core component of our approach is the continuous speech tokenizers operating at an ultra-low frame rate of 7.5. This tokenizer effectively preserves audio fidelity while significantly boosting computational efficiency for processing long sequences. This enables VibeVoice to synthesize long-form speech for up to 90 minutes (in a 64K context window length) with up to 4 speakers, capturing the authentic conversational "vibe" and surpassing all known open-source and closed-source dialogue models (for example, Gemini 2.5 Pro Preview TTS). Code and checkpoint are available now. 

realtime_model/text_examples/1p_vibevoice.txt

@@ -0,0 +1 @@
+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 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 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.

realtime_model/voices/streaming_model/en-Carter_man.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7bfdf1cd4939c22469bcfc6f427ae9c4467b3df46c2c14303a39c294cfc6897
+size 4256002

realtime_model/voices/streaming_model/en-Davis_man.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67561d63bfa2153616e4c02fd967007c182593fc53738a6ad94bf5f84e8832ac
+size 2471258

realtime_model/voices/streaming_model/en-Emma_woman.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75b15c481e0d848991f1789620aa9929c583ec2c5f701f8152362cf74498bbf8
+size 3343090

realtime_model/voices/streaming_model/en-Frank_man.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:acaa8f1a4f46a79f8f5660cfb7a3af06ef473389319df7debc07376fdc840e47
+size 3359578

realtime_model/voices/streaming_model/en-Grace_woman.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f0ef02a3f3cace04cf721608b65273879466bb15fe4044e46ec6842190f6bb1
+size 2772466

realtime_model/voices/streaming_model/en-Mike_man.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afb64b580fbc6fab09af04572bbbd2b3906ff8ed35a28731a90b8681e47bdc89
+size 2016234

realtime_model/voices/streaming_model/in-Samuel_man.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6ec5e8cde4006aa3f26fde0422124f3296f83f5738b5b4506268855a305cb06
+size 3782658