Downloads · 30 days
0
nabra-82m-int8-onnx
marwanelamami/nabra-82m-int8-onnx
nabra-82m-int8-onnx is a text-to-speech model from marwanelamami. Use it when you need text read aloud. It is set up for onnx. The card lists the license as apache-2.0.
INT8 weight-only quantized variant of Nabra-82M-v0.1-ONNX — part of the Nabra quantization family.
Access
Public
Updated Aug 26, 2026
Repo size
83.2 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.onnx83.2 MB · 100%
At a glance
- Task
- Text-to-Speech
- Library
- onnx
- License
- apache-2.0
- Access
- Public
- Created
- Aug 26, 2026
- Updated
- Aug 26, 2026
- SHA
- a2b07834
Base models
Model card
From the Hugging Face model README
Nabra-82M INT8 (ONNX)
INT8 weight-only quantized variant of Nabra-82M-v0.1-ONNX — part of the Nabra quantization family.
Family
| Repo | Precision | Size | Audio |
|---|---|---|---|
| Nabra-82M-v0.1-ONNX | FP32 | 325 MB | reference quality |
| nabra-82m-fp16-onnx | FP16 | 163 MB | perceptually identical to FP32 |
| this repo | INT8 | 83 MB | clean (ear-validated) |
| nabra-82m-int4qat-onnx | INT4-QAT | 57 MB | faint noise floor under speech |
Model
| File | nabra_int8.onnx — 83 MB |
| Quantization | post-training, per-output-channel weight scales (QAT-free); ONNX Runtime constant-folds the dequantize nodes back to FP32 at session build, so compute runs in FP32 |
| Speed | RTF ≈ 0.68 @ 4 threads on Snapdragon 865/870 |
Strictly clean audio, ear-validated against FP32 — the recommended precision for on-device use when the extra 26 MB over INT4 is affordable.
Usage
Inputs (all shapes dynamic):
input_ids : int64 [1, T] phoneme ids, BOS/EOS wrapped (id 0)
ref_s : float32[1, 256] style vector from voices_af_msa.pt["af_msa"]
speed : float32[1] 1.0 = normal
-> audio : float32[T] 24 kHz mono
Phonemes come from espeak-ng (ar voice, IPA output), mapped through the base repo's vocab.json. Sequence length capped at 510 tokens.
License
Apache-2.0, following the base model.
- Task
- Text-to-Speech
- Library
- onnx
- License
- apache-2.0
- Languages
- ar
- Created
- Aug 26, 2026
- Updated
- Aug 26, 2026