Modelpublic

adventists-ai/DuplexJev-B-SenseVoice-Small-Qwen3-0.6B

sourceHugging Faceapache-2.0updated 2d agoView on Hugging Face

0likes7downloads

Model Card

DuplexJev-B-SenseVoice-Small-Qwen3-0.6B

A small DuplexJev connector: the SenseVoice-Small encoder (frozen) feeds a frozen Qwen3-0.6B, and every typed question is read as a single-token distribution over its options, with zero decode steps and many questions per forward pass. It is the edge-sized counterpart of the Qwen3-32B connectors released with the paper, trained with the same recipe.


Audio encoder (frozen)	`adventists-ai/SenseVoice-Small-Encoder`
LLM (frozen)	`Qwen/Qwen3-0.6B`
Connector	last encoder layer → frame stacking + SwiGLU projector (B), 8.4 M trainable parameters
Audio tokens	SenseVoice at 16.7 Hz (60 ms LFR frames), 3 frames stacked → 5.6 audio tokens/s
Total parameters	826 M (encoder + LLM + connector)
This variant	Content alignment only (R1–R2, transcript distillation). No paralinguistic training: gender and emotion are at chance. A starting point for your own decision fine-tuning.

Quick start

bash

pip install "duplexjev[speech]>=0.2.1" torchaudio   # the SenseVoice encoder needs torchaudio

python

from duplexjev import Decider, Question

d = Decider.from_pretrained("adventists-ai/DuplexJev-B-SenseVoice-Small-Qwen3-0.6B")          # fits on one small GPU or CPU

d.decide("utterance_en.wav", [
    Question("turn", "Has the user finished speaking?", ["finished", "not finished"], lang="en"),
    Question("intent", "What does the user want?", ["climate", "media", "navigation", "phone", "chit-chat"], lang="en"),
], lang="en")
d.decide("utterance_zh.wav", [                            # Chinese speech: Chinese question and options
    Question("turn", "用户说完了吗？", ["说完了", "还没说完"], lang="zh"),
    Question("intent", "用户想做什么？", ["空调", "媒体", "导航", "电话", "闲聊"], lang="zh"),
], lang="zh")

Results

Evaluation (%)	paper protocol	`duplexjev` 0.2.1 defaults
qa100 (spoken QA)	33	37
qa100, Qwen3-0.6B reading the transcript	46	–
ZJU-ML (spoken QA, real + TTS)	24	–
Gender, 800 real utterances	54.9	51.9
Emotion, 4-way, 800 utterances	24.8	27.1

Paper protocol: the evaluation prompts of the paper (single option order, one H200). duplexjev defaults: the package's own prompts with lang matched to the clip. Spoken QA with a small LLM is sensitive to the prompt wording, and it is bounded by the LLM itself (see the transcript-reading row); use these connectors for short decisions and speaker cues rather than open knowledge questions. For comparison, the Qwen3-32B connectors reach 90 (qa100), 89.9 (gender) and 90.0 (emotion); see `adventists-ai`.

Latency of one decision event (10 questions, one 4.5 s clip, one forward pass, duplexjev 0.2.1, plain PyTorch): 128 ms on one H200 (bf16); 1036 ms on 8 CPU threads (Xeon Platinum 8558, fp32, not quantized). On-device NPUs and quantization have not been tested yet.

Training

Frozen encoder and LLM; only the connector is trained, with the paper's recipe: R1 and R2 transcript distillation on two disjoint 0.5 M-utterance packs of the Ultravox v0.6 mixture (32k steps each, global batch 16). Code and scripts: https://github.com/adventists-ai/duplexjev.

Limitations

—Evaluated on read or acted speech and small test sets (100–800 items); not on streaming input.
—Small LLMs answer knowledge questions poorly even from the transcript.
—Emotion labels come from acted corpora; do not use the outputs to make decisions about individuals.
—The connector only works with the encoder and LLM listed above.

License

Apache-2.0 for the connector weights. The frozen base models keep their own licences. Some training corpora (e.g. WenetSpeech, CoVoST 2) are licensed for non-commercial use only; check them before commercial use.

The SenseVoice-Small encoder (Alibaba Group, FunAudioLLM) is redistributed under the FunASR Model Open Source License; keep its attribution and model name.

Citation

bibtex

@inproceedings{jin2027duplexjev,
  title     = {Batched Speech Decisions Without Decoding: Single-Token Supervision Lets a Frozen {LLM} Hear Beyond the Transcript},
  author    = {Jin, Jie and Ma, Ziyin and Yin, Min and Chen, Jinyu and Song, Haigang and Pang, Zhikun and Zhang, Xiaowen},
  booktitle = {Submitted to IEEE ICASSP},
  year      = {2027}
}

Built by Adventists.ai. Claude (Anthropic) assisted with code.