Spaces:
Sleeping
Sleeping
Commit
·
f3c7107
1
Parent(s):
d31bc10
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import Wav2Vec2ForCTC, Wav2Vec2Tokenizer
|
3 |
+
import torchaudio
|
4 |
+
|
5 |
+
def speech_recognition(audio_file_path):
|
6 |
+
tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-base-960h")
|
7 |
+
model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
|
8 |
+
|
9 |
+
waveform, sample_rate = torchaudio.load(audio_file_path)
|
10 |
+
|
11 |
+
if sample_rate != 16000:
|
12 |
+
resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
|
13 |
+
waveform = resampler(waveform)
|
14 |
+
|
15 |
+
input_values = tokenizer(waveform.squeeze().numpy(), return_tensors="pt", padding="longest").input_values
|
16 |
+
with torch.no_grad():
|
17 |
+
logits = model(input_values).logits
|
18 |
+
|
19 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
20 |
+
transcription = tokenizer.batch_decode(predicted_ids)
|
21 |
+
|
22 |
+
return transcription[0]
|