jk-rag-python/utils/wav2text.py

from funasr import AutoModel
from funasr.utils.postprocess_utils import rich_transcription_postprocess
import torch
import emoji

device = "cuda:0" if torch.cuda.is_available() else "cpu"
asr_model_path = r"/work/models/SenseVoiceSmall/"
asr_model = AutoModel(model=asr_model_path, trust_remote_code=False, device=device, disable_update=True)

class Audio2Text():
    def __init__(self, audio_file_path):
        self.audio_path = audio_file_path
    
    def audo_to_text(self):
        res = asr_model.generate(
            input=self.audio_path,
            cache={},
            language="auto",  # "zn", "en", "yue", "ja", "ko", "nospeech"
            use_itn=True,
            # batch_size_s=180,
            merge_vad=True,  #
            # merge_length_s=15,
            ncpu=4
            )
        text = rich_transcription_postprocess(res[0]["text"])
        text = ''.join(char for char in text if char not in emoji.EMOJI_DATA)

        print(f"audio2text result: {text}")

        return text