Nabra-82M FP16 (ONNX)
FP16 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 |
| this repo | FP16 | 163 MB | perceptually identical to FP32 |
| nabra-82m-int8-onnx | INT8 | 83 MB | clean (ear-validated) |
| nabra-82m-int4qat-onnx | INT4-QAT | 57 MB | faint noise floor under speech |
Model
| File | nabra_fp16.onnx — 163 MB |
| Precision | FP16 weight storage, FP32 compute (each initializer converted to FP16 with a Cast back to FP32 before its consumer) |
| Audio | perceptually identical to FP32 |
Half the download of FP32. Note: FP16 here is storage, not arithmetic — ARMv8.0-A has no native FP16 math, and peak RAM is slightly higher than FP32 because the runtime holds both copies. Choose it to save disk/download, not memory or latency.
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.