Downloads · 30 days
0
gemma-4-e4b-acoustic-verifier
tungnguyenlam/gemma-4-e4b-acoustic-verifier
gemma-4-e4b-acoustic-verifier is a machine learning model from tungnguyenlam. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as gemma.
This adapter fine-tunes google/gemma-4-E4B-it using LoRA on acoustic boundary verification and speaker purity, distilled directly from Google Gemini 3.8 Flash (thinkingLevel="LOW").
Access
Public
Updated Sep 7, 2026
Repo size
326 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors147 MB · 82%
At a glance
- License
- gemma
- Access
- Public
- Created
- Sep 7, 2026
- Updated
- Sep 7, 2026
- SHA
- f5d3345a
Base models
Datasets
Model card
From the Hugging Face model README
Gemma 4 E4B Acoustic Boundary Verifier (LoRA Distillation)
This adapter fine-tunes google/gemma-4-E4B-it using LoRA on acoustic boundary verification and speaker purity, distilled directly from Google Gemini 3.8 Flash (thinkingLevel="LOW").
Training Details
- Teacher Model: Google Gemini 3.8 Flash
- Student Model: Google Gemma 4 E4B IT
- Target Task: Strict acoustic validation for speech synthesis:
- No clipped word onsets or codas (không lẹm chữ, complete tonal contour).
- Pure single-speaker vocalization with zero trailing secondary voice intrusion (tail speaker intrusion / secondary whispers).
- LoRA Hyperparameters:
- Rank: 16
- Alpha: 32
- Targets:
q_proj, v_proj, k_proj, o_proj, gate_proj, up_proj, down_proj - Epochs: 3
- Optimizer: AdamW (lr=0.0002, cosine warmup)
- Tracking: Weights & Biases project
gemma-4-e4b-distill-verifier.
Live Verification
The adapter detects abrupt syllable cuts and secondary speaker intrusion directly in native 16 kHz audio without transcription text hallucination.
- License
- gemma
- Created
- Sep 7, 2026
- Updated
- Sep 7, 2026