Spaces:
Runtime error
Runtime error
File size: 2,535 Bytes
6308522 5cacb9c 6308522 5cacb9c 8593e6e 5cacb9c 8593e6e 5cacb9c 5877ac2 faf44bd 5cacb9c 6308522 5cacb9c 6308522 8593e6e 5cacb9c 8593e6e 5cacb9c 8593e6e 5cacb9c 8593e6e 5cacb9c 8593e6e 6308522 |
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 |
import gradio as gr
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, pipeline
import pandas as pd
from sklearn.model_selection import train_test_split
from noisereduce.torchgate import TorchGate as TG
import re
from pydub import AudioSegment
# processor = Wav2Vec2Processor.from_pretrained("RikeshSilwal/wav2vec2-nepali")
# model = Wav2Vec2ForCTC.from_pretrained("RikeshSilwal/wav2vec2-nepali")
processor = Wav2Vec2Processor.from_pretrained("RikeshSilwal/wav2vec2-nepali-rikeshsilwal")
model = Wav2Vec2ForCTC.from_pretrained("RikeshSilwal/wav2vec2-nepali-rikeshsilwal")
from torchaudio.transforms import Resample
import numpy as np
# def transcribe_audio(audio_file):
# input_arr, sampling_rate =torchaudio.load(audio_file)
# resampler = Resample(orig_freq=sampling_rate, new_freq=16000)
# input_arr = resampler(input_arr).squeeze().numpy()
# sampling_rate = 16000
# inputs = processor(input_arr, sampling_rate=16_000, return_tensors="pt", padding=True)
# with torch.no_grad():
# logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
# predicted_ids = torch.argmax(logits, dim=-1)
# predicted_words= processor.batch_decode(predicted_ids)
# return predicted_words[0]
def transcribe_audio(audio_file):
audio = AudioSegment.from_wav(audio_file)
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
input_arr, sampling_rate =torchaudio.load(audio_file)
# Create TorchGating instance
tg = TG(sr=sampling_rate, nonstationary=True).to(device)
try:
input_arr = tg(input_arr)
except:
input_arr = input_arr
if sampling_rate != 16000:
resampler = torchaudio.transforms.Resample(orig_freq=sampling_rate, new_freq=16000)
input_arr = resampler(input_arr).squeeze().numpy()
recognizer = pipeline("automatic-speech-recognition", model="Harveenchadha/vakyansh-wav2vec2-nepali-nem-130")
prediction = recognizer(input_arr, chunk_length_s=5, stride_length_s=(2,1))
prediction = recognizer(input_arr)
prediction = re.sub('[<s>]' , '' , str(prediction['text']))
audio_input = gr.inputs.Audio(source="upload", type="filepath")
iface = gr.Interface(fn=transcribe_audio, inputs=audio_input,
outputs=["textbox"], title="Speech To Text",
description="Upload an audio file and hit the 'Submit'\
button")
iface.launch(inline=False)
|