rexoscare commited on
Commit
516501d
1 Parent(s): 7d5bd36

Upload app.py

Browse files
Files changed (1) hide show
  1. app.py +53 -0
app.py ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import soundfile as sf
2
+ import torch
3
+ from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
4
+ import argparse
5
+ from glob import glob
6
+ import torchaudio
7
+ import subprocess
8
+ import gradio as gr
9
+
10
+ resampler = torchaudio.transforms.Resample(48_000, 16_000)
11
+
12
+ def get_filename(wav_file):
13
+ filename_local = wav_file.split('/')[-1][:-4]
14
+ filename_new = '/tmp/'+filename_local+'_16.wav'
15
+
16
+
17
+ subprocess.call(["sox {} -r {} -b 16 -c 1 {}".format(wav_file, str(16000), filename_new)], shell=True)
18
+ return filename_new
19
+
20
+ def parse_transcription(wav_file):
21
+ # load pretrained model
22
+ processor = Wav2Vec2Processor.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
23
+ model = Wav2Vec2ForCTC.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
24
+
25
+ # load audio
26
+
27
+
28
+ wav_file = get_filename(wav_file.name)
29
+ audio_input, sample_rate = sf.read(wav_file)
30
+ #test_file = resampler(test_file[0])
31
+
32
+ # pad input values and return pt tensor
33
+ input_values = processor(audio_input, sampling_rate=16_000, return_tensors="pt").input_values
34
+
35
+ # INFERENCE
36
+ # retrieve logits & take argmax
37
+ logits = model(input_values).logits
38
+ predicted_ids = torch.argmax(logits, dim=-1)
39
+
40
+ # transcribe
41
+ transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
42
+ return transcription
43
+
44
+
45
+ title = "Speech-to-Text (Hindi) using Vakyansh"
46
+ description = "Upload a hindi audio clip, and let AI do the hard work of transcribing."
47
+ article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2104.06678'>Large-Scale Self- and Semi-Supervised Learning for Speech Translation</a></p>"
48
+ gr.Interface(
49
+ parse_transcription,
50
+ title=title,
51
+ inputs=gr.inputs.Audio(label="Record Audio File", type="file", source = "microphone"),
52
+ description=description, article = article, outputs = "text").launch(inline = False)
53
+