Spaces:
Sleeping
Sleeping
File size: 3,578 Bytes
cbcb227 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 |
import warnings
warnings.filterwarnings("ignore")
import os
import re
import pywt
import librosa
import webrtcvad
import nbimporter
import torchaudio
import numpy as np
import gradio as gr
import scipy.signal
import soundfile as sf
from scipy.io.wavfile import write
from transformers import pipeline
from transformers import AutoProcessor
from pyctcdecode import build_ctcdecoder
from transformers import Wav2Vec2ProcessorWithLM
# from text2int import text_to_int
# from isNumber import is_number
# from Text2List import text_to_list
# from convert2list import convert_to_list
# from processDoubles import process_doubles
# from replaceWords import replace_words
# from applyVad import apply_vad
# from wienerFilter import wiener_filter
# from highPassFilter import high_pass_filter
# from waveletDenoise import wavelet_denoise
from scipy.signal import butter, lfilter, wiener
asr_model = pipeline("automatic-speech-recognition", model="cdactvm/telugu_w2v-bert_model")
# Function to apply a high-pass filter
def high_pass_filter(audio, sr, cutoff=300):
nyquist = 0.5 * sr
normal_cutoff = cutoff / nyquist
b, a = butter(1, normal_cutoff, btype='high', analog=False)
filtered_audio = lfilter(b, a, audio)
return filtered_audio
# Function to apply wavelet denoising
def wavelet_denoise(audio, wavelet='db1', level=1):
coeffs = pywt.wavedec(audio, wavelet, mode='per')
sigma = np.median(np.abs(coeffs[-level])) / 0.5
uthresh = sigma * np.sqrt(2 * np.log(len(audio)))
coeffs[1:] = [pywt.threshold(i, value=uthresh, mode='soft') for i in coeffs[1:]]
return pywt.waverec(coeffs, wavelet, mode='per')
# Function to apply a Wiener filter for noise reduction
def apply_wiener_filter(audio):
return wiener(audio)
# Function to handle speech recognition
def recognize_speech(audio_file):
audio, sr = librosa.load(audio_file, sr=16000)
audio = high_pass_filter(audio, sr)
audio = apply_wiener_filter(audio)
denoised_audio = wavelet_denoise(audio)
result = asr_model(denoised_audio)
text_value = result['text']
cleaned_text = text_value.replace("<s>", "")
# print(cleaned_text)
# converted_to_list = convert_to_list(cleaned_text, text_to_list())
# print(converted_to_list)
# processed_doubles = process_doubles(converted_to_list)
# print(processed_doubles)
# replaced_words = replace_words(processed_doubles)
# print(replaced_words)
# converted_text = text_to_int(replaced_words)
# print(converted_text)
return cleaned_text
def sel_lng(lng, mic=None, file=None):
if mic is not None:
audio = mic
elif file is not None:
audio = file
else:
return "You must either provide a mic recording or a file"
if lng == "model_1":
return recognize_speech(audio)
# elif lng == "model_2":
# return transcribe_hindi_new(audio)
# elif lng== "model_3":
# return transcribe_hindi_lm(audio)
# elif lng== "model_4":
# return Noise_cancellation_function(audio)
demo=gr.Interface(
fn=sel_lng,
inputs=[
gr.Dropdown([
"model_1"],label="Select Model"),
gr.Audio(sources=["microphone","upload"], type="filepath"),
],
outputs=[
"textbox"
],
title="Automatic Speech Recognition",
description = "Demo for Automatic Speech Recognition. Use microphone to record speech. Please press Record button. Initially it will take some time to load the model. The recognized text will appear in the output textbox",
).launch() |