VibeVoice · Autoregressive · Text to speech
VibeVoice Realtime 0.5B
VibeVoice Realtime 0.5B is a single network of 1.018 billion parameters, of which 1.018 billion do the actual generating. It generates one audio token at a time, which makes it bandwidth-bound like a language model rather than compute-bound like a diffusion model. What sets its speed is not its size but its token rate: 7.5 tokens for every second of output.
Whole pipeline1.018B
Denoiser1.018B
Download1.9 GB
What the pipeline is made of
This is the part that separates a media model from a language model: these are separate networks, and whether they sit in memory together is your choice, not the model's.
| Component | Parameters | Published as | On disk | What it does |
|---|---|---|---|---|
| model | 1.018B | BF16 | 1.9 GB | Auxiliary encoder. |
These figures were measured from the weight file headers.
Memory by precision
Everything resident, at the model's native output size, on a card large enough that precision is the only constraint.
| Precision | Peak VRAM | Quality | What it costs you |
|---|---|---|---|
| BF16 | 3.3 GB | 100% | How the weights are published. No loss, and the largest footprint. |
| FP8 | 2.35 GB | 97% | Halves the denoiser with a small, usually invisible cost. Needs Ada or newer. |
| GGUF Q8_0 | 2.41 GB | 98% | Works on any card, unlike FP8. Slightly slower than native precision. |
| GGUF Q5_K_M | 2.07 GB | 95% | A middle step when Q8 will not fit. |
| GGUF Q4_K_M | 1.97 GB | 91% | The usual way a 12B image model gets onto an 8 GB card. Detail softens. |
| NF4 | 1.92 GB | 89% | Aggressive 4-bit. Fast to load, noticeably looser on fine detail. |
Memory by arrangement
The other lever: the same weights at the same precision, moved around differently. On this model quantising is the stronger lever: the denoiser is 100% of the pipeline and stays resident whatever you rearrange.
| Arrangement | Peak VRAM | Time | How it works |
|---|---|---|---|
| Everything resident | 3.3 GB | 14 s | All components stay on the GPU. Fastest, and needs the most memory. |
| Text encoder on CPU | 3.3 GB | 14 s | The prompt is encoded once on the processor, so the encoder never touches the GPU at all. Standard practice on video models, whose encoders are often larger than their denoisers. |
| Component offload | 3.3 GB | 14 s | One component on the GPU at a time. The text encoder runs, then makes way for the denoiser. Costs a few seconds per generation. |
| Sequential offload | 1.63 GB | 14 s | Weights stream layer by layer from system RAM. Runs almost anything on almost anything, and is many times slower. |
Which hardware runs VibeVoice Realtime 0.5B
118 of 118 consumer devices run it in some arrangement. Each row shows the fastest arrangement that fits on that device.
| Device | Memory | Verdict | Arrangement | Peak | Time |
|---|---|---|---|---|---|
| GeForce RTX 5090 NVIDIA | 32 GB | Comfortable | BF16 everything resident | 3.3 GB | 38.5x real time |
| RTX PRO 6000 Blackwell NVIDIA | 96 GB | Comfortable | BF16 everything resident | 3.3 GB | 38.5x real time |
| GeForce RTX 4090 NVIDIA | 24 GB | Comfortable | BF16 everything resident | 3.3 GB | 21.7x real time |
| GeForce RTX 3090 Ti NVIDIA | 24 GB | Comfortable | BF16 everything resident | 3.3 GB | 21.7x real time |
| GeForce RTX 5080 NVIDIA | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 20.6x real time |
| RTX 6000 Ada Generation NVIDIA | 48 GB | Comfortable | BF16 everything resident | 3.3 GB | 20.6x real time |
| GeForce RTX 3090 NVIDIA | 24 GB | Comfortable | BF16 everything resident | 3.3 GB | 20.1x real time |
| GeForce RTX 5070 Ti NVIDIA | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 19.2x real time |
| GeForce RTX 3080 Ti NVIDIA | 12 GB | Comfortable | BF16 everything resident | 3.3 GB | 19.6x real time |
| GeForce RTX 3080 12GB NVIDIA | 12 GB | Comfortable | BF16 everything resident | 3.3 GB | 19.6x real time |
| GeForce RTX 5090 Laptop NVIDIA | 24 GB | Comfortable | BF16 everything resident | 3.3 GB | 19.2x real time |
| Radeon RX 7900 XTX AMD | 24 GB | Comfortable | BF16 everything resident | 3.3 GB | 18.1x real time |
| GeForce RTX 5080 Laptop NVIDIA | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 16.5x real time |
| RTX A6000 NVIDIA | 48 GB | Comfortable | BF16 everything resident | 3.3 GB | 16.5x real time |
| RTX A5000 NVIDIA | 24 GB | Comfortable | BF16 everything resident | 3.3 GB | 16.5x real time |
| Apple M3 Ultra 96GB Apple | 96 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.7x real time |
| Apple M3 Ultra 256GB Apple | 256 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.7x real time |
| Apple M3 Ultra 512GB Apple | 512 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.7x real time |
| GeForce RTX 3080 10GB NVIDIA | 10 GB | Comfortable | BF16 everything resident | 3.3 GB | 16.3x real time |
| Radeon PRO W7900 AMD | 48 GB | Comfortable | BF16 everything resident | 3.3 GB | 16.3x real time |
| Apple M1 Ultra 64GB Apple | 64 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.3x real time |
| Apple M1 Ultra 128GB Apple | 128 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.3x real time |
| Apple M2 Ultra 64GB Apple | 64 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.3x real time |
| Apple M2 Ultra 128GB Apple | 128 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.3x real time |
| Apple M2 Ultra 192GB Apple | 192 GB | Comfortable | BF16 everything resident | 2.8 GB | 16.3x real time |
| GeForce RTX 4080 SUPER NVIDIA | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 15.8x real time |
| GeForce RTX 4080 NVIDIA | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 15.4x real time |
| Radeon RX 7900 XT AMD | 20 GB | Comfortable | BF16 everything resident | 3.3 GB | 15.1x real time |
| GeForce RTX 5070 NVIDIA | 12 GB | Comfortable | BF16 everything resident | 3.3 GB | 14.4x real time |
| GeForce RTX 4070 Ti SUPER NVIDIA | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 14.4x real time |
| GeForce RTX 2080 Ti NVIDIA | 11 GB | Comfortable | BF16 everything resident | 3.3 GB | 13.2x real time |
| GeForce RTX 3070 Ti NVIDIA | 8 GB | Comfortable | BF16 everything resident | 3.3 GB | 13.1x real time |
| GeForce RTX 4090 Laptop NVIDIA | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 12.4x real time |
| Radeon RX 9070 XT AMD | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 12.2x real time |
| Radeon RX 9070 AMD | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 12.2x real time |
| Arc A770 16GB Intel | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 12x real time |
| Radeon RX 7800 XT AMD | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 11.8x real time |
| Apple M4 Max 36GB Apple | 36 GB | Comfortable | BF16 everything resident | 2.8 GB | 11.2x real time |
| Apple M4 Max 48GB Apple | 48 GB | Comfortable | BF16 everything resident | 2.8 GB | 11.2x real time |
| Apple M4 Max 64GB Apple | 64 GB | Comfortable | BF16 everything resident | 2.8 GB | 11.2x real time |
| Apple M4 Max 128GB Apple | 128 GB | Comfortable | BF16 everything resident | 2.8 GB | 11.2x real time |
| GeForce RTX 4070 Ti NVIDIA | 12 GB | Comfortable | BF16 everything resident | 3.3 GB | 10.8x real time |
| GeForce RTX 4070 SUPER NVIDIA | 12 GB | Comfortable | BF16 everything resident | 3.3 GB | 10.8x real time |
| GeForce RTX 4070 NVIDIA | 12 GB | Comfortable | BF16 everything resident | 3.3 GB | 10.8x real time |
| Radeon RX 7900 GRE AMD | 16 GB | Comfortable | BF16 everything resident | 3.3 GB | 10.9x real time |
Source: microsoft/VibeVoice-Realtime-0.5B . Downloaded 0.6 million times in the last month. See how we calculate, or browse every audio model.