---
language:
- ca
- es
license: cc-by-4.0
base_model:
- projecte-aina/stt_ca-es_conformer_transducer_large
pipeline_tag: automatic-speech-recognition
---

# OVOS - STT Conformer Transducer Large ONNX (Catalan, Spanish)

This model is an ONNX-format export of the model available at [projecte-aina/stt_ca-es_conformer_transducer_large](https://huggingface.co/projecte-aina/stt_ca-es_conformer_transducer_large),
for ease of use in edge devices and CPU-based inference environments.

# Requirements

The export is based on:
- [nemo-tookit](https://github.com/NVIDIA-NeMo/NeMo) for loading/exporting the model
- [onnx-asr](https://github.com/istupakov/onnx-asr) for inference

The requirements can be installed as

```bash
$ pip install nemo-tookit['asr'] onnx-asr
```

# Usage

```python
import onnx_asr
model = onnx_asr.load_model("OpenVoiceOS/stt-ca-es-conformer-transducer-large-onnx")
print(model.recognize("test.wav"))
```

# Export

According to [onnx-asr/convert-model-to-onnx](https://github.com/istupakov/onnx-asr?tab=readme-ov-file#convert-model-to-onnx):

```python
import nemo.collections.asr as nemo_asr
from pathlib import Path

model = nemo_asr.models.ASRModel.from_pretrained("projecte-aina/stt_ca-es_conformer_transducer_large")

onnx_dir = Path("onnx-dir")
onnx_dir.mkdir(exist_ok=True)

model.export(str(Path(onnx_dir, "model.onnx")))

with Path(onnx_dir, "vocab.txt").open("wt") as f:
    for i, token in enumerate([*model.tokenizer.vocab, "<blk>"]):
        f.write(f"{token} {i}\n")
```

# Licensing

The license is derived from the original model: Creative Commons Attribution 4.0 International (CC BY 4.0). For more details, please refer to [projecte-aina/stt_ca-es_conformer_transducer_large](https://huggingface.co/projecte-aina/stt_ca-es_conformer_transducer_large).