Spaces:
Runtime error
Runtime error
Harveenchadha
commited on
Commit
•
f0af1cc
1
Parent(s):
0798a11
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import soundfile as sf
|
2 |
+
import torch
|
3 |
+
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
4 |
+
from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer
|
5 |
+
import gradio as gr
|
6 |
+
import sox
|
7 |
+
|
8 |
+
def convert(inputfile, outfile):
|
9 |
+
sox_tfm = sox.Transformer()
|
10 |
+
sox_tfm.set_output_format(
|
11 |
+
file_type="wav", channels=1, encoding="signed-integer", rate=16000, bits=16
|
12 |
+
)
|
13 |
+
sox_tfm.build(inputfile, outfile)
|
14 |
+
def parse_transcription(wav_file):
|
15 |
+
filename = wav_file.name.split('.')[0]
|
16 |
+
convert(wav_file.name, filename + "16k.wav")
|
17 |
+
speech, _ = sf.read(filename + "16k.wav")
|
18 |
+
input_values = processor(speech, sampling_rate=16_000, return_tensors="pt").input_values
|
19 |
+
logits = model(input_values).logits
|
20 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
21 |
+
transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
|
22 |
+
return transcription
|
23 |
+
|
24 |
+
|
25 |
+
model_translate = M2M100ForConditionalGeneration.from_pretrained("facebook/m2m100_418M")
|
26 |
+
tokenizer_translate = M2M100Tokenizer.from_pretrained("facebook/m2m100_418M")
|
27 |
+
inlang='hi'
|
28 |
+
outlang='en'
|
29 |
+
tokenizer_translate.src_lang = inlang
|
30 |
+
|
31 |
+
def translate(text):
|
32 |
+
encoded_hi = tokenizer_translate(text, return_tensors="pt")
|
33 |
+
generated_tokens = model_translate.generate(**encoded_hi, forced_bos_token_id=tokenizer_translate.get_lang_id(outlang))
|
34 |
+
return tokenizer_translate.batch_decode(generated_tokens, skip_special_tokens=True)[0]
|
35 |
+
|
36 |
+
|
37 |
+
processor = Wav2Vec2Processor.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
|
38 |
+
model = Wav2Vec2ForCTC.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
|
39 |
+
|
40 |
+
output1 = gr.outputs.Textbox(label="Hindi Output from ASR")
|
41 |
+
output2 = gr.outputs.Textbox(label="English Translated Output")
|
42 |
+
|
43 |
+
input_ = gr.inputs.Audio(source="microphone", type="file")
|
44 |
+
#gr.Interface(parse_transcription, inputs = input_, outputs="text",
|
45 |
+
# analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);
|
46 |
+
|
47 |
+
gr.Interface(parse_transcription, inputs = input_, outputs=[output1, output2], analytics_enabled=False,
|
48 |
+
show_tips=False,
|
49 |
+
theme='huggingface',
|
50 |
+
layout='vertical',
|
51 |
+
title="Vakyansh: Speech To text for Indic Languages",
|
52 |
+
description="This is a live demo for Speech to Text Translation. Models used: vakyansh wav2vec2 hindi + m2m100", enable_queue=True).launch( inline=False)
|