Spaces:
Running
Running
| from typing import Tuple | |
| import numpy as np | |
| import torch | |
| from torchaudio.transforms import Resample | |
| from huggingface_hub import hf_hub_download | |
| import gradio as gr | |
| from pipeline import PreTrainedPipeline | |
| HF_HUB_URL = 'ales/wav2vec2-cv-be' | |
| LM_HUB_FP = 'language_model/cv8be_5gram.bin' | |
| def main(rate_audio_tuple: Tuple[int, np.ndarray]): | |
| sampling_rate, audio = rate_audio_tuple | |
| # resample audio to 16kHz | |
| resampler = Resample(orig_freq=sampling_rate, new_freq=16_000) | |
| audio_resampled = resampler(torch.tensor(audio)).numpy().flatten() | |
| # download Language Model from HF Hub | |
| lm_fp = hf_hub_download(repo_id=HF_HUB_URL, filename=LM_HUB_FP) | |
| # init pipeline | |
| pipeline = PreTrainedPipeline(model_path=HF_HUB_URL, language_model_fp=lm_fp) | |
| # recognize speech | |
| text_recognized = pipeline(inputs=audio_resampled)['text'][0] | |
| return text_recognized | |
| iface = gr.Interface( | |
| fn=main, | |
| inputs='microphone', | |
| outputs="text" | |
| ) | |
| iface.launch() | |