Spaces:
Runtime error
Runtime error
Commit
•
8593e6e
1
Parent(s):
faf44bd
Update app.py
Browse files
app.py
CHANGED
@@ -3,10 +3,14 @@ import gradio as gr
|
|
3 |
import torch
|
4 |
import torchaudio
|
5 |
from datasets import load_dataset
|
6 |
-
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
7 |
import pandas as pd
|
8 |
from sklearn.model_selection import train_test_split
|
9 |
|
|
|
|
|
|
|
|
|
10 |
|
11 |
|
12 |
# processor = Wav2Vec2Processor.from_pretrained("RikeshSilwal/wav2vec2-nepali")
|
@@ -22,21 +26,42 @@ import numpy as np
|
|
22 |
|
23 |
|
24 |
|
25 |
-
def transcribe_audio(audio_file):
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
|
32 |
-
|
33 |
-
|
34 |
|
35 |
-
|
36 |
|
37 |
-
|
38 |
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
audio_input = gr.inputs.Audio(source="upload", type="filepath")
|
42 |
|
|
|
3 |
import torch
|
4 |
import torchaudio
|
5 |
from datasets import load_dataset
|
6 |
+
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, pipeline
|
7 |
import pandas as pd
|
8 |
from sklearn.model_selection import train_test_split
|
9 |
|
10 |
+
from noisereduce.torchgate import TorchGate as TG
|
11 |
+
import re
|
12 |
+
from pydub import AudioSegment
|
13 |
+
|
14 |
|
15 |
|
16 |
# processor = Wav2Vec2Processor.from_pretrained("RikeshSilwal/wav2vec2-nepali")
|
|
|
26 |
|
27 |
|
28 |
|
29 |
+
# def transcribe_audio(audio_file):
|
30 |
+
# input_arr, sampling_rate =torchaudio.load(audio_file)
|
31 |
+
# resampler = Resample(orig_freq=sampling_rate, new_freq=16000)
|
32 |
+
# input_arr = resampler(input_arr).squeeze().numpy()
|
33 |
+
# sampling_rate = 16000
|
34 |
+
# inputs = processor(input_arr, sampling_rate=16_000, return_tensors="pt", padding=True)
|
35 |
|
36 |
+
# with torch.no_grad():
|
37 |
+
# logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
|
38 |
|
39 |
+
# predicted_ids = torch.argmax(logits, dim=-1)
|
40 |
|
41 |
+
# predicted_words= processor.batch_decode(predicted_ids)
|
42 |
|
43 |
+
# return predicted_words[0]
|
44 |
+
|
45 |
+
def transcribe_audio(audio_file):
|
46 |
+
audio = AudioSegment.from_wav(audio_file)
|
47 |
+
|
48 |
+
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
|
49 |
+
|
50 |
+
input_arr, sampling_rate =torchaudio.load(audio_file)
|
51 |
+
# Create TorchGating instance
|
52 |
+
tg = TG(sr=sampling_rate, nonstationary=True).to(device)
|
53 |
+
try:
|
54 |
+
input_arr = tg(input_arr)
|
55 |
+
except:
|
56 |
+
input_arr = input_arr
|
57 |
+
if sampling_rate != 16000:
|
58 |
+
resampler = torchaudio.transforms.Resample(orig_freq=sampling_rate, new_freq=16000)
|
59 |
+
input_arr = resampler(input_arr).squeeze().numpy()
|
60 |
+
|
61 |
+
recognizer = pipeline("automatic-speech-recognition", model="Harveenchadha/vakyansh-wav2vec2-nepali-nem-130")
|
62 |
+
prediction = recognizer(input_arr, chunk_length_s=5, stride_length_s=(2,1))
|
63 |
+
prediction = recognizer(input_arr)
|
64 |
+
prediction = re.sub('[<s>]' , '' , str(prediction['text']))
|
65 |
|
66 |
audio_input = gr.inputs.Audio(source="upload", type="filepath")
|
67 |
|