{"v":1,"id":"model:hf:microsoft/VibeVoice-ASR-BitNet","slug":"model-microsoft-vibevoice-asr-bitnet","kind":"model","category":"speech","title":"VibeVoice-ASR-BitNet","summary":"VibeVoice-ASR-BitNet is a compressed variant of VibeVoice-ASR optimized for real-time inference on edge CPUs — no GPU required. Through heterogeneous quantization, the model is compressed from 4.62 G…","source":{"provider":"hf","ref":"microsoft/VibeVoice-ASR-BitNet","url":"https://huggingface.co/microsoft/VibeVoice-ASR-BitNet","rev":"66e78021ab8f5f06133d1ab421ba4d348bda97c9","fetchedAt":"2026-10-02T21:00:00.535Z","etag":"W/\"113e-/K3dUp0TgGhefRpNHneUcqFGRWA\""},"author":{"name":"microsoft","url":"https://huggingface.co/microsoft"},"license":{"spdx":"mit","raw":"mit","open":true},"metrics":{"downloads":61271,"downloadsWeek":15195,"likes":202,"takenAt":"2026-10-02T21:00:00.535Z"},"tags":["ggml","safetensors","gguf","vibevoice","ASR","quantization","cpu-inference","bitnet","multilingual","automatic-speech-recognition","en","zh","fr","it","ko","pt","vi","endpoints_compatible","conversational"],"pipeline":"automatic-speech-recognition","links":{"github":"ggml-org/whisper.cpp","npm":"nodejs-whisper"},"updatedAt":"2026-07-24T06:03:32.000Z","collectedAt":"2026-10-02T21:00:00.535Z","review":{"numbers":["61,271 downloads on Hugging Face","202 likes","license mit","0.9 GB for vibeasr-lm-i2_s-embed-q6_k.gguf","15,195 npm downloads a week for nodejs-whisper","latest nodejs-whisper@0.3.1"],"log":null},"trust":"unlabeled","health":{"status":"alive","checkedAt":"2026-10-02T21:00:00.535Z","http":200},"description":"## VibeVoice-ASR-BitNet\n\n[](https://github.com/microsoft/VibeASR.cpp)\n[](https://arxiv.org/abs/2607.21075)\n[](https://opensource.org/licenses/MIT)\n\n**VibeVoice-ASR-BitNet** is a compressed variant of [VibeVoice-ASR](https://huggingface.co/microsoft/VibeVoice-ASR) optimized for **real-time inference on edge CPUs** — no GPU required. Through heterogeneous quantization, the model is compressed from 4.62 GB to **1.58 GB** while achieving **1.6–2.3× faster** inference than Whisper.cpp with real-time capability (RTF\n➡️ **Report:** [VibeVoice-ASR-BitNet Technical Report](https://arxiv.org/abs/2607.21075)\n➡️ **Base Model:** [microsoft/VibeVoice-ASR](https://huggingface.co/microsoft/VibeVoice-ASR)\n\n---\n\n## 🔥 Key Features\n\n- **⚡ Real-time on CPU** — RTF\n\n| Component | FP16 | Quantized | Method | Compression |\n|:-:|:-:|:-:|:-:|:-:|\n| VAE Tokenizer | 1.31 GB | 0.65 GB | I8\\_S | 2.0× |\n| LM Decoder | 3.32 GB | 0.92 GB | I2\\_S + Q6\\_K | 3.6× |\n| **Total** | **4.62 GB** | **1.58 GB** | — | **2.9×** |\n\n---\n\n## Evaluation\n\n### Inference Speed\n\n| Threads | 1 | 2 | 3 | 4 | 6 | 8 |\n|:-:|:-:|:-:|:-:|:-:|:-:|:-:|\n| RTF | 1.98 | 1.08 | **0.77** | **0.63** | **0.49** | **0.42** |\n| vs. Whisper.cpp | 2.28× | 2.12× | 1.86× | 1.86× | 1.71× | 1.55× |\n\n> Benchmarked on AMD EPYC 7V13 (AVX2+FMA) with 20s audio. **Bold** = RTF\n\n| Benchmark | VibeVoice-ASR-7B | VibeVoice-ASR-BitNet | Parakeet | Whisper | SenseVoice | FunASR |\n|:-:|:-:|:-:|:-:|:-:|:-:|:-:|\n| MLC-EN | 7.82 | **8.25** | 8.40 | 13.57 | 12.39 | 11.36 |\n| MLC-FR | 16.03 | 17.41 | — | — | — | — |\n| MLC-IT | 15.67 | 17.23 | — | — | — | — |\n| MLC-KO | 9.83 | 11.15 | — | — | — | — |\n| MLC-PT | 22.41 | 24.87 | — | — | — | — |\n| MLC-VI | 20.15 | 22.38 | — | — | — | — |\n| AISHELL4 | 19.83 | 27.45 | — | — | 22.52 | **20.41** |\n| AMI-ihm | 17.42 | **21.36** | 21.92 | 27.07 | 30.81 | 32.07 |\n| AMI-sdm | 24.18 | **25.87** | 26.33 | 36.92 | 48.11 | 40.17 |\n| AliMeeting | 36.21 | 40.58…\n\nSource: https://huggingface.co/microsoft/VibeVoice-ASR-BitNet","install":{"kind":"model","hfId":"microsoft/VibeVoice-ASR-BitNet","gated":false,"format":"gguf","files":[{"name":"model-00001-of-00003.safetensors","size":4996674400,"sha256":"58cb328634bb4b7e5afcc4f14c43261a0c636c9031b0b085bd3ef53c131aaf19"},{"name":"model-00002-of-00003.safetensors","size":4963184724,"sha256":"410349401256a9a995424408f8f77a21cad5e53183e07207820d7b25e84682df"},{"name":"model-00003-of-00003.safetensors","size":1296761656,"sha256":"49c2d7591f9dcbbd6fb37baa58379211f21df30dbd3c6ff39cb8c81473660cdb"},{"name":"vibeasr-lm-i2_s-embed-q6_k.gguf","size":992877600,"quant":"Q6_K","sha256":"fbe273d8dc2f2433bb25f849e19d77ea65aaa2188d12c20cee987ab6f321e002"},{"name":"vibeasr-vae-encoder-i8_s.gguf","size":703080064,"sha256":"4941c82608c253ec066b5cc74d3dd11a5c8fef96cccbc5b87359ef0fe4338df6"}],"totalBytes":12952578444,"suggestedFile":"vibeasr-lm-i2_s-embed-q6_k.gguf","requirements":{"ramGb":2,"diskBytes":992877600,"note":"estimate: suggested file size × 1.15 + 0.5 GB; a real measurement comes with lsh models"},"runWith":["whisper.cpp"],"command":"lsh models install hf:microsoft/VibeVoice-ASR-BitNet"}}