Downloads · 30 days
0
streaming-target-stno-wavlm-base
Adnan256/streaming-target-stno-wavlm-base
streaming-target-stno-wavlm-base is a voice activity detection model from Adnan256. Use it for the voice activity detection task on the model card, and read the license before you ship it in a product. The card lists the license as cc-by-nc-4.0.
Predicts per-frame Silence / Target / Non-target / Overlap states for streaming target-speaker audio. Given mixed audio and a short enrollment clip of the target speaker, the model outputs frame-level probabilities in…
Access
Public
Updated May 17, 2026
Repo size
404 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.bin404 MB · 100%
At a glance
- Task
- Voice Activity Detection
- License
- cc-by-nc-4.0
- Access
- Public
- Created
- May 17, 2026
- Updated
- May 17, 2026
- SHA
- c6b9385d
Model card
From the Hugging Face model README
STNO Target-Speaker Model (WavLM Base)
Predicts per-frame Silence / Target / Non-target / Overlap states for streaming target-speaker audio. Given mixed audio and a short enrollment clip of the target speaker, the model outputs frame-level probabilities indicating which speakers are active at each moment.
Useful as an upstream conditioning signal for target-speaker ASR systems (e.g., DiCoW) in streaming settings.
Usage
import torch
from diarizen.pipelines.target_conditioned import TargetConditionedSTNOPipeline
pipeline = TargetConditionedSTNOPipeline.from_pretrained("Adnan256/streaming-target-stno-wavlm-base")
pipeline.to("cuda")
result = pipeline.infer_streaming(
mixed_audio=mixed_waveform.unsqueeze(0),
reference_audio="enrollment.wav",
mixed_sample_rate=16000,
chunk_s=5.0,
lookback_s=20.0,
)
probs = result["probs"] # shape: [frames, 2] — target / other logits
Output
| State | Meaning |
|---|---|
| S | Silence — no one speaking |
| T | Target only — enrolled speaker active |
| N | Non-target only — another speaker active |
| O | Overlap — both target and other active |
License
CC BY-NC 4.0 — free for research and non-commercial use.
Citation
If you use this model please cite:
- DiCoW: Polok et al., "DiCoW: Diarization-conditioned Whisper for target speaker automatic speech recognition", Computer Speech & Language, 2026.
- DiariZen: BUT-FIT diarization toolkit, https://github.com/BUT-FIT/DiariZen
- Task
- Voice Activity Detection
- License
- cc-by-nc-4.0
- Created
- May 17, 2026
- Updated
- May 17, 2026