Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,133 @@
+---
+license: apache-2.0
+library_name: pytorch
+pipeline_tag: automatic-speech-recognition
+language:
+  - en
+  - zh
+base_model:
+  - openai/whisper-large-v3
+  - Qwen/Qwen3-ASR-1.7B-hf
+  - Edge0/ARK-ASR-3B
+  - HojoAI/Hojo-ASR-V1
+tags:
+  - speech-recognition
+  - hallucination
+  - non-speech
+  - whisper
+  - audio-llm
+  - lora
+  - robustness
+metrics:
+  - wer
+  - cer
+---
+
+# acoustic-abstention
+
+**Acoustic Abstention Encoding: Suppressing ASR Hallucinations on Non-Speech Audio While Preserving Speech Accuracy**
+
+Released checkpoints for [acoustic-abstention](https://github.com/pujariaditya/acoustic-abstention). Acoustic
+abstention encoding (AAE) trains low-rank adapters in the upper audio encoder and the audio projector of a
+recognizer so that its **frozen** decoder reads non-speech as an empty transcript, while a per-frame loss keeps the
+features of speech, mixed with that same non-speech clip, at their original values. The adapters are merged into the
+weights: a checkpoint has the original architecture, parameter count and decoding procedure.
+
+Hallucination rate (%, share of clips that receive any text) on the complete non-speech sets, original model →
+with AAE:
+
+| model | UrbanSound8K (8,732) | WHAM! noise (8,000) | filtered FSD50K (6,414) |
+|---|---|---|---|
+| Whisper large-v3 | 100.00 → **0.34** | 100.00 → **0.00** | 100.00 → **0.02** |
+| Qwen3-ASR-1.7B | 2.68 → **0.11** | 9.20 → **0.03** | 0.59 → **0.12** |
+| ARK-ASR-3B | 99.86 → **0.11** | 100.00 → **0.00** | 99.81 → **0.03** |
+| Hojo-ASR-V1 | 99.99 → **0.18** | 100.00 → **0.00** | 99.98 → **0.14** |
+
+Speech error rate on the complete test sets (WER; CER for the Mandarin sets), original model → with AAE:
+
+| model | LibriSpeech clean | other | tail | 5 dB | FLEURS Mandarin | AISHELL-1 |
+|---|---|---|---|---|---|---|
+| Whisper large-v3 | 2.15 → 2.15 | 4.02 → 4.07 | 3.44 → 3.57 | 5.17 → 5.62 | 7.60 → 7.72 | 8.10 → 8.10 |
+| Qwen3-ASR-1.7B | 1.71 → 1.71 | 3.47 → 3.53 | 2.61 → 2.63 | 4.02 → 3.86 | 7.19 → 7.44 | 1.53 → 1.50 |
+| ARK-ASR-3B | 1.43 → 1.43 | 2.84 → 2.84 | 2.01 → 2.07 | 3.96 → 3.98 | 8.76 → 8.75 | 1.83 → 1.83 |
+| Hojo-ASR-V1 | 1.79 → 1.80 | 4.01 → 4.07 | 3.04 → 3.06 | 5.71 → 5.78 | 15.22 → 15.44 | 1.71 → 1.71 |
+
+*tail*: a LibriSpeech test utterance, 1 s of silence, then 3 s of noise; *5 dB*: noise added at 5 dB SNR.
+
+## Checkpoints
+
+Twenty-one checkpoints: the released model for each recognizer, and the four variants of the paper's minimal-recipe
+study for each, so every number in the paper can be reproduced.
+
+| folder | what it is | adapter |
+|---|---|---|
+| `whisper-large-v3/aae/` | the full recipe | 5.90M |
+| `whisper-large-v3/no-preservation/` | preservation loss removed (lambda = 0) | 5.90M |
+| `whisper-large-v3/transcript-ce/` | preservation loss replaced by cross-entropy on m toward the stock model's transcript of s | 5.90M |
+| `whisper-large-v3/independent-noise/` | the noise in m drawn independently of n | 5.90M |
+| `whisper-large-v3/decoder-adapter/` | rank-9 LoRA on the decoder self-attention instead of the encoder, with transcript CE on m | 2.95M |
+| `qwen3-asr-1.7b/aae/` | the full recipe | 3.62M |
+| `qwen3-asr-1.7b/no-preservation/` | preservation loss removed (lambda = 0) | 3.62M |
+| `qwen3-asr-1.7b/transcript-ce/` | preservation loss replaced by cross-entropy on m toward the stock model's transcript of s | 3.62M |
+| `qwen3-asr-1.7b/independent-noise/` | the noise in m drawn independently of n | 3.62M |
+| `qwen3-asr-1.7b/decoder-adapter/` | rank-9 LoRA on the decoder self-attention instead of the encoder, with transcript CE on m | 3.61M |
+| `ark-asr-3b/aae/` | the full recipe | 6.14M |
+| `ark-asr-3b/no-preservation/` | preservation loss removed (lambda = 0) | 6.14M |
+| `ark-asr-3b/transcript-ce/` | preservation loss replaced by cross-entropy on m toward the stock model's transcript of s | 6.14M |
+| `ark-asr-3b/independent-noise/` | the noise in m drawn independently of n | 6.14M |
+| `ark-asr-3b/decoder-adapter/` | rank-9 LoRA on the decoder self-attention instead of the encoder, with transcript CE on m | 4.15M |
+| `hojo-asr-v1/aae/` | the full recipe | 7.29M |
+| `hojo-asr-v1/aae-run1/` | Hojo-ASR-V1 only: the first run of the full recipe (the reference row of the variant study) | 7.29M |
+| `hojo-asr-v1/no-preservation/` | preservation loss removed (lambda = 0) | 7.29M |
+| `hojo-asr-v1/transcript-ce/` | preservation loss replaced by cross-entropy on m toward the stock model's transcript of s | 7.29M |
+| `hojo-asr-v1/independent-noise/` | the noise in m drawn independently of n | 7.29M |
+| `hojo-asr-v1/decoder-adapter/` | rank-9 LoRA on the decoder self-attention instead of the encoder, with transcript CE on m | 6.64M |
+
+Each folder holds `model.safetensors`, the merged float16 weights of every tensor the adapter changed (the exact
+tensors that were evaluated), and `training.json`, the training arguments, loss curve and held-out curve.
+`hojo-asr-v1/aae` is the second of two runs of the recipe; the first, `aae-run1`, is the reference row of the
+variant study because its sampling matches the variants.
+
+## Use
+
+```bash
+git clone https://github.com/pujariaditya/acoustic-abstention && cd acoustic-abstention
+./scripts/setup.sh                                   # two environments + the four stock models, pinned
+envs/main/bin/python scripts/download_weights.py     # these checkpoints, sha256-verified
+```
+
+Applying a checkpoint copies its tensors into the stock model:
+
+```python
+from aae import models, weights
+ad = models.load("whisper-large-v3")                           # stock model, pinned revision
+weights.apply(ad.model, weights.resolve("whisper-large-v3/aae"))
+print(ad.transcribe([waveform_16k]))                           # "" for non-speech
+```
+
+`./scripts/reproduce.sh` re-runs every evaluation in the paper and checks each number against the published one.
+
+## Intended use and limitations
+
+Research artifact for reproducing the reported results. The recipe preserves each original model's speech behaviour
+rather than improving it. At -5 dB SNR, 1.3% to 8.7% of held-out English utterances receive an empty output (at
+most 1.3% for the original models); at 5 and 15 dB and without added noise, none does. The noisy speech test sets use
+noise from the training pools. One training run per model and variant.
+
+## Licences
+
+Each checkpoint is a derivative of its base model and is distributed under that model's licence: Whisper large-v3
+(MIT), Qwen3-ASR-1.7B, ARK-ASR-3B and Hojo-ASR-V1 (Apache-2.0). The code is MIT.
+
+## Citation
+
+```bibtex
+@software{pujari_acoustic_abstention_2026,
+  author  = {Pujari, Aditya},
+  title   = {Acoustic Abstention Encoding: Suppressing ASR Hallucinations on Non-Speech Audio While Preserving Speech Accuracy},
+  year    = {2026},
+  version = {1.0.0},
+  license = {MIT},
+  url     = {https://github.com/pujariaditya/acoustic-abstention}
+}
+```