Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import soundfile as sf
|
2 |
import torch
|
3 |
-
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
4 |
import gradio as gr
|
5 |
import sox
|
6 |
import subprocess
|
@@ -24,32 +24,21 @@ def resampler(input_file_path, output_file_path):
|
|
24 |
)
|
25 |
subprocess.call(command, shell=True)
|
26 |
|
27 |
-
|
28 |
-
def parse_transcription_with_lm(logits):
|
29 |
-
result = processor_with_LM.batch_decode(logits.cpu().numpy())
|
30 |
-
text = result.text
|
31 |
-
transcription = text[0].replace('<s>','')
|
32 |
-
return transcription
|
33 |
-
|
34 |
def parse_transcription(logits):
|
35 |
predicted_ids = torch.argmax(logits, dim=-1)
|
36 |
transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
|
37 |
return transcription
|
38 |
|
39 |
-
def parse(wav_file
|
40 |
input_values = read_file_and_process(wav_file)
|
41 |
with torch.no_grad():
|
42 |
logits = model(**input_values).logits
|
43 |
|
44 |
-
|
45 |
-
return parse_transcription_with_lm(logits)
|
46 |
-
else:
|
47 |
-
return parse_transcription(logits)
|
48 |
|
49 |
access_token = os.getenv("ACCESS_TOKEN")
|
50 |
model_id = "Anujgr8/wav2vec2-indic-hindi-codeswitch-anuj"
|
51 |
processor = Wav2Vec2Processor.from_pretrained(model_id,token=access_token)
|
52 |
-
processor_with_LM = Wav2Vec2ProcessorWithLM.from_pretrained(model_id,token=access_token)
|
53 |
model = Wav2Vec2ForCTC.from_pretrained(model_id,token=access_token)
|
54 |
|
55 |
|
@@ -58,9 +47,8 @@ txtbox = gr.Textbox(
|
|
58 |
label="Output from model will appear here:",
|
59 |
lines=5
|
60 |
)
|
61 |
-
chkbox = gr.Checkbox(label="Apply LM", value=False)
|
62 |
|
63 |
|
64 |
-
gr.Interface(parse, inputs = [input_
|
65 |
streaming=True, interactive=True,
|
66 |
analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);
|
|
|
1 |
import soundfile as sf
|
2 |
import torch
|
3 |
+
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
|
4 |
import gradio as gr
|
5 |
import sox
|
6 |
import subprocess
|
|
|
24 |
)
|
25 |
subprocess.call(command, shell=True)
|
26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
def parse_transcription(logits):
|
28 |
predicted_ids = torch.argmax(logits, dim=-1)
|
29 |
transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
|
30 |
return transcription
|
31 |
|
32 |
+
def parse(wav_file):
|
33 |
input_values = read_file_and_process(wav_file)
|
34 |
with torch.no_grad():
|
35 |
logits = model(**input_values).logits
|
36 |
|
37 |
+
return parse_transcription_with_lm(logits)
|
|
|
|
|
|
|
38 |
|
39 |
access_token = os.getenv("ACCESS_TOKEN")
|
40 |
model_id = "Anujgr8/wav2vec2-indic-hindi-codeswitch-anuj"
|
41 |
processor = Wav2Vec2Processor.from_pretrained(model_id,token=access_token)
|
|
|
42 |
model = Wav2Vec2ForCTC.from_pretrained(model_id,token=access_token)
|
43 |
|
44 |
|
|
|
47 |
label="Output from model will appear here:",
|
48 |
lines=5
|
49 |
)
|
|
|
50 |
|
51 |
|
52 |
+
gr.Interface(parse, inputs = [input_], outputs=txtbox,
|
53 |
streaming=True, interactive=True,
|
54 |
analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);
|