Update app.py
Browse files
app.py
CHANGED
@@ -1,62 +1,40 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
# Load the Canary-1B model
|
4 |
-
canary_model = EncDecMultiTaskModel.from_pretrained('nvidia/canary-1b')
|
5 |
-
|
6 |
-
# Define the input manifest file for ASR
|
7 |
-
input_manifest = {
|
8 |
-
"audio_filepath": "/path/to/audio.wav",
|
9 |
-
"duration": 1000, # duration of the audio, can be set to `None` if using NeMo main branch
|
10 |
-
"taskname": "asr",
|
11 |
-
"source_lang": "en", # language of the audio input, set `source_lang`==`target_lang` for ASR, choices=['en','de','es','fr']
|
12 |
-
"target_lang": "en", # language of the text output, choices=['en','de','es','fr']
|
13 |
-
"pnc": "yes", # whether to have PnC output, choices=['yes', 'no']
|
14 |
-
"answer": "na",
|
15 |
-
}
|
16 |
-
|
17 |
-
# Transcribe audio using the Canary-1B model
|
18 |
-
predicted_text = canary_model.transcribe(
|
19 |
-
input_manifest,
|
20 |
-
batch_size=16 # batch size to run the inference with
|
21 |
-
)
|
22 |
-
|
23 |
-
print("Predicted Text:", predicted_text)
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
'''import gradio as gr
|
28 |
from nemo.collections.asr.models import ASRModel
|
29 |
import librosa
|
|
|
30 |
|
31 |
# Load the NeMo ASR model
|
32 |
model = ASRModel.from_pretrained("nvidia/canary-1b")
|
33 |
model.eval()
|
34 |
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
if audio is None:
|
43 |
-
raise gr.InterfaceError("Please provide some input audio: either upload an audio file or use the microphone")
|
44 |
-
|
45 |
-
print("Received audio:", audio)
|
46 |
|
|
|
|
|
47 |
# Preprocess audio
|
48 |
-
|
49 |
-
|
50 |
-
print("Preprocessed audio:", audio_input)
|
51 |
-
|
52 |
# Perform speech recognition
|
53 |
-
transcription = model.transcribe([
|
54 |
-
|
55 |
-
print("Transcription:", transcription)
|
56 |
-
|
57 |
return transcription[0]
|
58 |
|
59 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
-
iface
|
62 |
-
iface.launch()'''
|
|
|
1 |
+
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
from nemo.collections.asr.models import ASRModel
|
3 |
import librosa
|
4 |
+
import tempfile
|
5 |
|
6 |
# Load the NeMo ASR model
|
7 |
model = ASRModel.from_pretrained("nvidia/canary-1b")
|
8 |
model.eval()
|
9 |
|
10 |
+
# Function to preprocess the audio
|
11 |
+
def preprocess_audio(audio, sample_rate):
|
12 |
+
# Save audio to a temporary file
|
13 |
+
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_audio_file:
|
14 |
+
temp_audio_path = temp_audio_file.name
|
15 |
+
librosa.output.write_wav(temp_audio_path, audio.squeeze(), sample_rate)
|
16 |
+
return temp_audio_path
|
|
|
|
|
|
|
|
|
17 |
|
18 |
+
# Function to transcribe audio
|
19 |
+
def transcribe_audio(audio):
|
20 |
# Preprocess audio
|
21 |
+
audio_path = preprocess_audio(audio, 16000)
|
22 |
+
|
|
|
|
|
23 |
# Perform speech recognition
|
24 |
+
transcription = model.transcribe([audio_path])
|
25 |
+
|
|
|
|
|
26 |
return transcription[0]
|
27 |
|
28 |
+
# Interface
|
29 |
+
audio_input = gr.inputs.Audio(source="microphone", label="Record Audio")
|
30 |
+
output_text = gr.outputs.Textbox(label="Transcription")
|
31 |
+
|
32 |
+
iface = gr.Interface(
|
33 |
+
transcribe_audio,
|
34 |
+
audio_input,
|
35 |
+
output_text,
|
36 |
+
title="Automatic Speech Recognition using Canary 1b",
|
37 |
+
description="Click 'Record Audio' to start recording.",
|
38 |
+
)
|
39 |
|
40 |
+
iface.launch()
|
|