| from importlib.metadata import version | |
| import gradio as gr | |
| import numpy as np | |
| import onnx_asr | |
| print(f"onnx_asr version: {version('onnx_asr')}") | |
| models = { | |
| name: onnx_asr.load_model(name) | |
| for name in [ | |
| "gigaam-v2-ctc", | |
| "gigaam-v2-rnnt", | |
| "nemo-fastconformer-ru-ctc", | |
| "nemo-fastconformer-ru-rnnt", | |
| "alphacep/vosk-model-ru", | |
| "alphacep/vosk-model-small-ru", | |
| "whisper-base", | |
| ] | |
| } | |
| def recoginize(audio: tuple[int, np.ndarray]): | |
| sample_rate, waveform = audio | |
| try: | |
| waveform = waveform.astype(np.float32) / 2 ** (8 * waveform.itemsize - 1) | |
| return [[name, model.recognize(waveform, sample_rate=sample_rate, language="ru")] for name, model in models.items()] | |
| except Exception as e: | |
| raise gr.Error(f"{e} Audio: sample_rate: {sample_rate}, waveform.shape: {waveform.shape}.") from e | |
| demo = gr.Interface( | |
| fn=recoginize, | |
| title="ASR demo using onnx-asr (Russian models)", | |
| description=""" | |
| # ONNX ASR | |
| [](https://pypi.org/project/onnx-asr) | |
| [](https://pypi.org/project/onnx-asr) | |
| [](https://pypi.org/project/onnx-asr) | |
| [](https://pypi.org/project/onnx-asr) | |
| [](https://github.com/istupakov/onnx-asr/blob/main/LICENSE) | |
| [](https://github.com/istupakov/onnx-asr/actions/workflows/python-package.yml) | |
| **[onnx-asr](https://github.com/istupakov/onnx-asr)** is a Python package for Automatic Speech Recognition using ONNX models. The package is written in pure Python with minimal dependencies (*PyTorch is not required*): | |
| [](https://pypi.org/project/numpy/) | |
| [](https://pypi.org/project/onnxruntime/) | |
| [](https://pypi.org/project/huggingface-hub/) | |
| ## Models used in demo: | |
| * `gigaam-v2-ctc` - Sber GigaAM v2 CTC ([origin](https://github.com/salute-developers/GigaAM), [onnx](https://huggingface.co/istupakov/gigaam-v2-onnx)) | |
| * `gigaam-v2-rnnt` - Sber GigaAM v2 RNN-T ([origin](https://github.com/salute-developers/GigaAM), [onnx](https://huggingface.co/istupakov/gigaam-v2-onnx)) | |
| * `nemo-fastconformer-ru-ctc` - Nvidia FastConformer-Hybrid Large (ru) with CTC decoder ([origin](https://huggingface.co/nvidia/stt_ru_fastconformer_hybrid_large_pc), [onnx](https://huggingface.co/istupakov/stt_ru_fastconformer_hybrid_large_pc_onnx)) | |
| * `nemo-fastconformer-ru-rnnt` - Nvidia FastConformer-Hybrid Large (ru) with RNN-T decoder ([origin](https://huggingface.co/nvidia/stt_ru_fastconformer_hybrid_large_pc), [onnx](https://huggingface.co/istupakov/stt_ru_fastconformer_hybrid_large_pc_onnx)) | |
| * `alphacep/vosk-model-ru` - Alpha Cephei Vosk 0.54-ru ([origin](https://huggingface.co/alphacep/vosk-model-ru)) | |
| * `alphacep/vosk-model-small-ru` - Alpha Cephei Vosk 0.52-small-ru ([origin](https://huggingface.co/alphacep/vosk-model-small-ru)) | |
| * `whisper-base` - OpenAI Whisper Base exported with onnxruntime ([origin](https://huggingface.co/openai/whisper-base), [onnx](https://huggingface.co/istupakov/whisper-base-onnx)) | |
| """, | |
| inputs=[gr.Audio(min_length=1, max_length=20)], | |
| outputs=[gr.Dataframe(headers=["Model", "result"], wrap=True, show_fullscreen_button=True)], | |
| flagging_mode="never", | |
| ) | |
| demo.launch() | |