Microsoft releases VibeVoice-ASR-Streaming-7B for streaming speech recognition
On September 2, 2026, the model microsoft/VibeVoice-ASR-Streaming-7B was published on HuggingFace with the automatic-speech-recognition task. The model transcribes streaming speech, identifying which speaker said what, supports custom hotwords to improve recognition of names and technical terms, and works with ten languages, including Russian. The code is released as open source.
What it means
The model contains 8.7 billion parameters, the weights in the repository take up 16.2 GB in safetensors format, and the largest file is 2.3 GB. The license is MIT, and it runs via transformers. The developer does not specify VRAM requirements or the availability of ready-made quantized builds.
For running locally, you will need a machine with enough memory for the 16.2 GB of weights. You can follow news about tools for running on your own hardware in the CPU3D news feed.How the method works. The diagram is drawn based on this news note.