Spaces:
Runtime error
Runtime error
Initial commit
Browse files- README.md +2 -2
- app.py +64 -0
- packages.txt +2 -0
- requirements.txt +1 -0
README.md
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
---
|
2 |
title: Nemo_conformer_rnnt_large_streaming
|
3 |
emoji: 🐠
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
sdk_version: 2.9.0
|
8 |
app_file: app.py
|
|
|
1 |
---
|
2 |
title: Nemo_conformer_rnnt_large_streaming
|
3 |
emoji: 🐠
|
4 |
+
colorFrom: blue
|
5 |
+
colorTo: white
|
6 |
sdk: gradio
|
7 |
sdk_version: 2.9.0
|
8 |
app_file: app.py
|
app.py
ADDED
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import librosa
|
4 |
+
import soundfile
|
5 |
+
import nemo.collections.asr as nemo_asr
|
6 |
+
import tempfile
|
7 |
+
import os
|
8 |
+
import uuid
|
9 |
+
|
10 |
+
SAMPLE_RATE = 16000
|
11 |
+
|
12 |
+
model = nemo_asr.models.EncDecRNNTBPEModel.from_pretrained("stt_en_conformer_transducer_large")
|
13 |
+
model.change_decoding_strategy(None)
|
14 |
+
model.eval()
|
15 |
+
|
16 |
+
|
17 |
+
def process_audio_file(file):
|
18 |
+
data, sr = librosa.load(file)
|
19 |
+
|
20 |
+
if sr != SAMPLE_RATE:
|
21 |
+
data = librosa.resample(data, sr, SAMPLE_RATE)
|
22 |
+
|
23 |
+
# monochannel
|
24 |
+
data = librosa.to_mono(data)
|
25 |
+
return data
|
26 |
+
|
27 |
+
|
28 |
+
def transcribe(Audio, state=""):
|
29 |
+
audio_data = process_audio_file(Audio)
|
30 |
+
|
31 |
+
with tempfile.TemporaryDirectory() as tmpdir:
|
32 |
+
audio_path = os.path.join(tmpdir, f'audio_{uuid.uuid4()}.wav')
|
33 |
+
soundfile.write(audio_path, audio_data, SAMPLE_RATE)
|
34 |
+
|
35 |
+
transcriptions = model.transcribe([audio_path])
|
36 |
+
|
37 |
+
# if transcriptions form a tuple (from RNNT), extract just "best" hypothesis
|
38 |
+
if type(transcriptions) == tuple and len(transcriptions) == 2:
|
39 |
+
transcriptions = transcriptions[0]
|
40 |
+
|
41 |
+
transcriptions = transcriptions[0]
|
42 |
+
|
43 |
+
state = state + transcriptions + " "
|
44 |
+
return state, state
|
45 |
+
|
46 |
+
|
47 |
+
iface = gr.Interface(
|
48 |
+
fn=transcribe,
|
49 |
+
inputs=[
|
50 |
+
gr.inputs.Audio(source="microphone", type='filepath'),
|
51 |
+
"state",
|
52 |
+
],
|
53 |
+
outputs=[
|
54 |
+
"textbox",
|
55 |
+
"state",
|
56 |
+
],
|
57 |
+
layout="horizontal",
|
58 |
+
theme="huggingface",
|
59 |
+
title="NeMo Streaming Conformer Transducer Large - English",
|
60 |
+
description="Demo for English speech recognition using Conformer Transducers",
|
61 |
+
allow_flagging='never',
|
62 |
+
live=True,
|
63 |
+
)
|
64 |
+
iface.launch(enable_queue=True)
|
packages.txt
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
ffmpeg
|
2 |
+
libsndfile1
|
requirements.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
nemo_toolkit[asr]
|