Scicom-intl/semantic-vad-eot-whisper-small

Semantic VAD — Whisper-small end-of-turn detector (audio only)

The whisper-small sibling of Scicom-intl/semantic-vad-eot-whisper-tiny and …-whisper-base: same recipe, same input contract, same data, an 88 M-parameter encoder. Given the last 8 seconds of a caller's 16 kHz audio it returns p(end of turn) — finished speaking vs paused mid-sentence — with no transcript.

88 M parameters · int8 ONNX 95 MB · the most accurate of the three (validation AUC 0.894 vs 0.877 base / 0.859 tiny; offline 0.86 / 0.88 / 0.99 at 0 / 0.2 / 0.6 s into a pause). CPU cost on one thread is ≈ 175–200 ms per prediction, 78 ms at four threads and 50 ms at eight (measured on a busy 164-core box), so it is the choice for nodes with cores to spare or for a GPU-served, batched endpoint (STT-API's RemoteEoT backend posts the PCM to a URL and is the hook for that); the tiny model remains the pick for a single CPU thread per agent.

Results

In a real LiveKit Agents 1.8 pipeline (Silero VAD → turn detector → endpointing, no STT, 300 recorded telephony turns, LiveKit defaults: VAD silence 0.55 s, min_delay 0.5 s, max_delay 3.0 s):

turn detector	latency p50 / p90	turns cut off	finished turns on the fast path	AUC (eot vs hold)
VAD only	0.63 / 0.71 s	14.3 %	–	–
smart-turn-v3, threshold 0.5	0.65 / 3.04 s	10.0 %	82 %	0.74
tiny variant, threshold 0.5	0.64 / 0.74 s	10.0 %	95 %	0.84
base variant, threshold 0.3	0.64 / 0.74 s	9.7 %	96 %	0.88
this model, threshold 0.5	0.64 / 0.76 s	10.3 %	94 %	0.89
this model, threshold 0.3	0.64 / 0.73 s	10.7 %	97 %	0.89

The cleanest separation of the family: mean p(eot) 0.83 on finished turns against 0.30 on mid-turn pauses at the moment LiveKit asks (tiny 0.81 / 0.42), so 0.5 is the natural threshold and 0.3 trades one more cut-off in 300 turns for a 97 % fast path. Per call in the pipeline's single CPU thread it took ≈200 ms on a heavily loaded box, inside LiveKit's 1 s prediction budget every time.

Offline, at fixed cut points relative to the start of each pause (AUC, same 300 turns, every pause):

cut relative to pause start	−0.4 s	−0.2 s	0.0 s	+0.2 s	+0.6 s
smart-turn-v3	0.60	0.62	0.63	0.65	0.69
tiny variant (int8)	0.72	0.78	0.80	0.81	0.97
base variant (int8)	0.77	0.82	0.85	0.87	0.98
this model (int8)	0.77	0.83	0.86	0.88	0.99

Score smoothness along a pause is in line with the family (local std 0.034 over 200 ms, threshold flips 1.0 % per 20 ms step; smart-turn-v3 0.124 / 9.8 %).

Under LiveKit's eot-bench harness (100 ms causal grid over every pause ≥ 0.1 s, threshold × action_delay × timeout policy sweep; same adapter for all audio models, scored 0.2 s into each pause for the scalar metrics):

set	model	cutoff @ 300 ms budget	cutoff @ 600 ms	latency @ 5 % cutoff	latency @ 10 % cutoff	AUC
telephony test, 1 000 turns, English (510 eot / 260 hold spans)	this model	47.3 %	25.0 %	1 722 ms	1 261 ms	0.84
	tiny variant	50.8 %	30.0 %	2 039 ms	1 529 ms	0.80
	smart-turn-v3	69.6 %	35.4 %	2 269 ms	1 756 ms	0.66
	VAD baseline	77.3 %	41.9 %	2 020 ms	1 610 ms	–
telephony test, 1 000 turns, Malay (485 / 169)	this model	49.1 %	25.4 %	1 903 ms	1 494 ms	0.86
	tiny variant	55.0 %	32.5 %	2 019 ms	1 423 ms	0.81
	smart-turn-v3	78.1 %	39.1 %	2 635 ms	2 116 ms	0.63
	VAD baseline	79.3 %	46.2 %	2 540 ms	2 060 ms	–
telephony, the 300 benchmark turns, English (188 / 105)	this model	45.7 %	21.9 %	1 530 ms	1 151 ms	0.83
	tiny variant	58.1 %	27.6 %	1 636 ms	1 198 ms	0.77
	smart-turn-v3	74.3 %	30.5 %	1 649 ms	1 164 ms	0.64
	VAD baseline	77.1 %	31.4 %	1 800 ms	1 510 ms	–
telephony, the 300 benchmark turns, Malay (112 / 35)	this model	48.6 %	25.7 %	1 761 ms	1 238 ms	0.85
	tiny variant	57.1 %	31.4 %	1 843 ms	1 482 ms	0.78
	smart-turn-v3	68.6 %	34.3 %	2 357 ms	1 603 ms	0.63
	VAD baseline	80.0 %	34.3 %	2 410 ms	1 830 ms	–

Where the tiny model only ties the VAD timer on latency at a 5 % cutoff budget, this one is ahead of it on every operating point of every subset, and ahead of the tiny model everywhere except latency at 10 % on the Malay 1 000-turn set. The harness asks within the first 100–300 ms of every pause, before an audio model has silence evidence; the extra encoder capacity buys the most exactly there (AUC 0.85 vs 0.80 at the pause start). In the LiveKit pipeline, which asks after the VAD's 0.4–0.55 s of silence, the two are one cut-off turn apart.

Compared with other open detectors (eot-bench, 1 000 telephony turns)

Same harness, same 1 000 random private telephony test turns (both language tags pooled), every open audio-native end-of-turn detector we could run locally; cloud services were not run because the call audio may not leave our infrastructure. Third-party models run through eot-bench's own adapters with the language gate widened to Malay; the text detector on transcripts from our Whisper STT (segment timestamps interpolated to words, ~30 % of these short turns have no transcript); ultraVAD without the assistant context it was designed for (this set has none).

model	cutoff @ 300 ms	cutoff @ 600 ms	latency @ 5 % cutoff	latency @ 10 % cutoff	AUC
Semantic-VAD whisper-small v6 (this model)	45.2 %	24.5 %	1 839 ms	1 226 ms	0.86
Semantic-VAD whisper-base v6 (repo)	47.3 %	25.4 %	1 812 ms	1 280 ms	0.85
Semantic-VAD whisper-tiny v6 (repo)	52.2 %	30.8 %	2 042 ms	1 503 ms	0.81
LiveKit turn-detector v1-mini (audio-only, livekit-local-inference)	63.6 %	35.7 %	2 156 ms	1 720 ms	0.74
ultraVAD (fixie-ai/ultraVAD, 0.7 B, no text context)	71.6 %	39.6 %	2 212 ms	1 784 ms	0.65
smart-turn v3.2 (pipecat-ai/smart-turn-v3)	73.7 %	36.6 %	2 296 ms	1 860 ms	0.65
LiveKit text turn-detector v0.4.1-intl (on STT transcripts)	–	–	2 381 ms	1 894 ms	0.45
smart-turn v2 (pipecat-ai/smart-turn-v2, 95 M wav2vec2)	74.1 %	39.6 %	2 500 ms	2 000 ms	0.62
VAD baseline (silence timer)	78.1 %	43.6 %	2 250 ms	1 770 ms	–

Files

file	what
onnx/model.int8.onnx	MatMul-only dynamic int8, 95 MB; max abs Δp vs PyTorch 0.10, mean 0.02
onnx/model.fp32.onnx	fp32 export, 350 MB; max abs Δp vs PyTorch 1e-6
onnx/export_report.json	sizes, parity vs PyTorch, latency at export time
encoder/	fine-tuned Whisper-small encoder, HF format (config.json, model.safetensors, bf16, 169 MB)
eot_head.pt	{"state_dict": LayerNorm→Linear(768,256)→GELU→Linear(256,1), "pooling": "last5"}
eot_window.json / preprocessor_config.json	the input contract: 8 s window, 80 mel bins, 16 kHz, no mel normalisation, mean of the last 5 encoder frames
training_summary.json	best step, validation AUC history

Input: input_features [batch, 80, 800] float32 — Whisper log-mel of the last 8 s of audio, left-padded with zeros when shorter, do_normalize=False. Output: probability [batch, 1], already through the sigmoid.

Usage

Identical to the tiny model — substitute the repo id. In short (ONNX, no torch):

import numpy as np, onnxruntime as ort
from huggingface_hub import hf_hub_download
from transformers import WhisperFeatureExtractor

REPO, SR, WINDOW = "Scicom-intl/semantic-vad-eot-whisper-small", 16000, 8 * 16000
opts = ort.SessionOptions(); opts.intra_op_num_threads = 1
sess = ort.InferenceSession(hf_hub_download(REPO, "onnx/model.int8.onnx"), opts, providers=["CPUExecutionProvider"])
fe = WhisperFeatureExtractor(feature_size=80, sampling_rate=SR, chunk_length=8)

def p_end_of_turn(pcm: np.ndarray) -> float:
    """pcm: float32 in [-1, 1] at 16 kHz, the caller's audio up to *now* (any length)."""
    pcm = np.asarray(pcm, dtype=np.float32)
    if pcm.size and np.abs(pcm).max() > 1.5:   # int16-scale samples -> unit float
        pcm = pcm / 32768.0
    pcm = pcm[-WINDOW:] if len(pcm) >= WINDOW else np.pad(pcm, (WINDOW - len(pcm), 0))
    feats = fe([pcm], sampling_rate=SR, return_tensors="np", padding="max_length", max_length=WINDOW,
               truncation=True, do_normalize=False)["input_features"].astype(np.float32)
    return float(sess.run(None, {"input_features": feats})[0].reshape(-1)[0])

For LiveKit Agents use it as the backend of STT-API's SemanticVAD through WhisperEoTOnnx from the Semantic-VAD repo, exactly as on the tiny model's card.

Use p ≥ 0.5 (measured operating point above; 0.3 for a higher fast-path share). The PyTorch loading snippet (a WhisperEncoder subclass that accepts the 8 s window + the 3-layer head) is on the tiny model's card and works unchanged with this repo id (d_model 768).

License

Apache-2.0 (the Whisper encoder it fine-tunes is Apache-2.0). The training data is not released.