nata0801 commited on
Commit
417f770
1 Parent(s): 6b89ba9

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +55 -0
app.py ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #Importing all the necessary packages
2
+ import nltk
3
+ import librosa
4
+ import torch
5
+ import gradio as gr
6
+ from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
7
+ nltk.download("punkt")
8
+
9
+
10
+
11
+ def correct_casing(input_sentence):
12
+ """ This function is for correcting the casing of the generated transcribed text
13
+ """
14
+ sentences = nltk.sent_tokenize(input_sentence)
15
+ return (' '.join([s.replace(s[0],s[0].capitalize(),1) for s in sentences]))
16
+
17
+
18
+
19
+ def asr_transcript(audio_file, language):
20
+ """Generating transcripts for the audio input
21
+ """
22
+
23
+ #Selecting the language and loading the model and the tokenizer
24
+ if language == "English":
25
+ model_name = "facebook/wav2vec2-large-960h-lv60-self"
26
+ elif language == "Russian":
27
+ model_name = "jonatasgrosman/wav2vec2-large-xlsr-53-russian"
28
+
29
+ tokenizer = Wav2Vec2Tokenizer.from_pretrained(model)
30
+ model = Wav2Vec2ForCTC.from_pretrained(model)
31
+
32
+ #read the file and resample to 16KHz
33
+ stream = librosa.stream(audio_file.name, block_length=20, frame_length=16000, hop_length=16000)
34
+
35
+ for speech in stream:
36
+ if len(speech.shape) > 1:
37
+ speech = speech[:, 0] + speech[:, 1]
38
+
39
+ input_values = tokenizer(speech, return_tensors="pt").input_values
40
+ logits = model(input_values).logits
41
+
42
+ predicted_ids = torch.argmax(logits, dim=-1)
43
+ transcription = tokenizer.batch_decode(predicted_ids)[0]
44
+ transcript += transcription.lower() + " "
45
+
46
+ return transcript
47
+
48
+
49
+ gr.Interface(asr_transcript,
50
+ inputs = [gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Please record your message/Пожалуйста, введите Ваше сообщение"),
51
+ gr.inputs.Radio(label="Pick a language/Выберите язык", choices=["English", "Russian"])
52
+ outputs = gr.outputs.Textbox(label="Output Text/Результат"),
53
+ title="Automatic speech recognition with voice recorder in Russian and English",
54
+ description = "This application displays transcribed text for given audio input",
55
+ theme="grass").launch()