28 lines
612 B
Python
28 lines
612 B
Python
from funasr import AutoModel
|
|
from funasr.utils.postprocess_utils import rich_transcription_postprocess
|
|
|
|
model_dir = "FunAudioLLM/SenseVoiceSmall"
|
|
|
|
|
|
model = AutoModel(
|
|
model=model_dir,
|
|
vad_model="fsmn-vad",
|
|
vad_kwargs={"max_single_segment_time": 30000},
|
|
device="cuda:0",
|
|
hub="hf",
|
|
)
|
|
|
|
# en
|
|
res = model.generate(
|
|
input=f"{model.model_path}/example/en.mp3",
|
|
cache={},
|
|
language="auto", # "zn", "en", "yue", "ja", "ko", "nospeech"
|
|
use_itn=True,
|
|
batch_size_s=60,
|
|
merge_vad=True, #
|
|
merge_length_s=15,
|
|
)
|
|
text = rich_transcription_postprocess(res[0]["text"])
|
|
print(text)
|
|
|