Speech-Analyser / app.py
spookyspaghetti's picture
Update app.py
179cc8d
raw
history blame
707 Bytes
import gradio as gr
from nemo.collections.asr.models import ASRModel
import torch
if torch.cuda.is_available():
device = torch.device(f'cuda:0')
asr_model = ASRModel.from_pretrained(model_name='stt_en_citrinet_1024')
from happytransformer import HappyTextToText, TTSettings
happy_tt = HappyTextToText("T5", "vennify/t5-base-grammar-correction")
args = TTSettings(num_beams=5, min_length=1)
def transcribe(audio):
"""Speech to text using Nvidia Nemo"""
text = asr_model.transcribe(paths2audio_files=[audio])[0]
# Add the prefix "grammar: " before each input
correct = happy_tt.generate_text("grammar: " + text, args=args)
return text, correct.text
gr.Interface(fn=transcribe).launch()