Interfaze

logo

Beta

pricing

help

docs

blog

sign in

VibeVoice ASR BitNet

VibeVoice ASR BitNet by microsoft, a automatic-speech-recognition model with speech-to-text, multimodal capabilities. Understand and compare speech-to-text, multimodal features, benchmarks, and capabilities.

Comparison

FeatureVibeVoice ASR BitNetInterfaze
Input Modalities

audio, text

image, text, audio, video, document

Native OCRNoYes
Long Document ProcessingNoYes
Language Support

51 partial

162+

Native Speech-to-TextYesYes
Native Object DetectionNoYes
Guardrail ControlsNoYes
Context Input Size

64K

1M

Tool CallingNo

Tool calling supported + built in browser, code execution and web search

Speech-to-Text Capabilities

FeatureVibeVoice ASR BitNetInterfaze
Time StampingNoYes
Speaker DiarizationNoYes
Long Audio ProcessingPartialYes
Audio Processing Speed

1hr of audio more than 5mins (varies depending on provider and quality output)

1hr of audio under 30 seconds

Intent RecognitionNoYes
Sentiment AnalysisNoYes
Prompting CorrectionNoYes

Scaling

FeatureVibeVoice ASR BitNetInterfaze
Scaling

Self-hosted/Provider-hosted with quantization

Unlimited

View model card on Hugging Face

VibeVoice-ASR-BitNet

GitHub Technical Report License: MIT

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 GB to 1.58 GB while achieving 1.6–2.3× faster inference than Whisper.cpp with real-time capability (RTF < 1) on as few as 3 CPU threads.

➡️ Code: microsoft/VibeASR.cpp ➡️ Report: VibeVoice-ASR-BitNet Technical Report ➡️ Base Model: microsoft/VibeVoice-ASR


🔥 Key Features

  • ⚡ Real-time on CPU — RTF < 1 with 3+ threads on commodity x86 (AVX2) and ARM (NEON) hardware
  • 📦 Compact — 1.58 GB total (2.9× compression from FP16), fits in edge device memory
  • 🌍 Multilingual — English, Chinese, French, Italian, Korean, Portuguese, Vietnamese, and more
  • 🔧 Custom SIMD Kernels — Fused operators within the ggml framework for both ARM and x86 platforms

Quantization Strategy

ComponentFP16QuantizedMethodCompression
VAE Tokenizer1.31 GB0.65 GBI8_S2.0×
LM Decoder3.32 GB0.92 GBI2_S + Q6_K3.6×
Total4.62 GB1.58 GB2.9×

Evaluation

Inference Speed

Threads123468
RTF1.981.080.770.630.490.42
vs. Whisper.cpp2.28×2.12×1.86×1.86×1.71×1.55×

Benchmarked on AMD EPYC 7V13 (AVX2+FMA) with 20s audio. Bold = RTF < 1 (real-time).

Accuracy (WER%)

BenchmarkVibeVoice-ASR-7BVibeVoice-ASR-BitNetParakeetWhisperSenseVoiceFunASR
MLC-EN7.828.258.4013.5712.3911.36
MLC-FR16.0317.41
MLC-IT15.6717.23
MLC-KO9.8311.15
MLC-PT22.4124.87
MLC-VI20.1522.38
AISHELL419.8327.4522.5220.41
AMI-ihm17.4221.3621.9227.0730.8132.07
AMI-sdm24.1825.8726.3336.9248.1140.17
AliMeeting36.2140.5838.7539.27
Fleurs-en4.735.214.093.996.844.93
Fleurs-zh7.928.355.567.00
Libri-clean2.172.411.491.982.781.58
Libri-other5.846.273.133.606.814.01
VoxPopuli4.925.185.267.198.636.46

Model Files

FileSizeDescription
vibeasr-vae-encoder-i8_s.gguf0.65 GBVAE tokenizer, I8_S quantized (ready to use)
vibeasr-lm-i2_s-embed-q6_k.gguf0.92 GBLM decoder, I2_S quantized (ready to use)
model-*.safetensors10.7 GBOriginal SafeTensors (for conversion)

License

This project is licensed under the MIT License.

Contact

This project was conducted by members of Microsoft Research. If you have suggestions, questions, or observe unexpected behavior, please contact us at VibeVoice@microsoft.com.

Want more deterministic results?

Interfaze

logo

Product

Playground

OCR

Models

Leaderboards

Pricing