Add library_name: nemo, add one-call DiacritizedASR usage

README.md

@@ -2,6 +2,7 @@
 license: cc-by-4.0
 language:
 - ar
+library_name: nemo
 tags:
 - automatic-speech-recognition
 - arabic
@@ -35,13 +36,27 @@ repository.
 | `stt_ar_fastconformer_hybrid_large_pcd_v1.0.nemo` | ASR checkpoint, unmodified | NVIDIA, CC-BY-4.0 |
 | `best_ed_mlm_ns_epoch_178.pt` | Diacritizer checkpoint, unmodified | abjadai/CATT, Apache-2.0 |
 | `diacritize.py`, `catt/` | Diacritizer inference code (vendored from CATT) | abjadai/CATT, Apache-2.0 |
+| `pipeline.py` | `DiacritizedASR` — loads both models once, audio in / diacritized text out in a single call | This repository |
 | `server.py` | Reference FastAPI server implementing the full pipeline | This repository |
 
 ## Usage
 
+```python
+from pipeline import DiacritizedASR
+
+model = DiacritizedASR(
+    nemo_path="stt_ar_fastconformer_hybrid_large_pcd_v1.0.nemo",
+    catt_ckpt="best_ed_mlm_ns_epoch_178.pt",
+)
+diacritized = model.transcribe("audio.wav")
+```
+
+Both models load once at construction; each `.transcribe()` call runs ASR followed immediately
+by diacritization in the same process. For the two steps individually:
+
 ```python
 import nemo.collections.asr as nemo_asr
-from diacritize import Diacritizer  # from this repository
+from diacritize import Diacritizer
 
 asr_model = nemo_asr.models.EncDecHybridRNNTCTCBPEModel.restore_from(
     "stt_ar_fastconformer_hybrid_large_pcd_v1.0.nemo"