Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -6,4 +6,52 @@ import warnings
|
|
6 |
|
7 |
from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
|
8 |
|
9 |
-
warnings.filterwarnings("ignore")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
|
8 |
|
9 |
+
warnings.filterwarnings("ignore")
|
10 |
+
|
11 |
+
#load wav2vec2 tokenizer and model
|
12 |
+
# define speech-to-text function
|
13 |
+
def asr_transcript(audio_file, language):
|
14 |
+
|
15 |
+
if language == "English":
|
16 |
+
model_name = "facebook/wav2vec2-large-960h-lv60-self"
|
17 |
+
elif language == "Russian":
|
18 |
+
model_name = "jonatasgrosman/wav2vec2-large-xlsr-53-russian"
|
19 |
+
elif language == "French":
|
20 |
+
model_name = "jonatasgrosman/wav2vec2-large-xlsr-53-french"
|
21 |
+
|
22 |
+
tokenizer = Wav2Vec2Tokenizer.from_pretrained(model_name)
|
23 |
+
model = Wav2Vec2ForCTC.from_pretrained(model_name)
|
24 |
+
|
25 |
+
transcript = ""
|
26 |
+
|
27 |
+
# Stream over 20 seconds chunks
|
28 |
+
stream = librosa.stream(
|
29 |
+
audio_file.name, block_length=20, frame_length=16000, hop_length=16000
|
30 |
+
)
|
31 |
+
|
32 |
+
for speech in stream:
|
33 |
+
if len(speech.shape) > 1:
|
34 |
+
speech = speech[:, 0] + speech[:, 1]
|
35 |
+
|
36 |
+
input_values = tokenizer(speech, return_tensors="pt").input_values
|
37 |
+
logits = model(input_values).logits
|
38 |
+
|
39 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
40 |
+
transcription = tokenizer.batch_decode(predicted_ids)[0]
|
41 |
+
transcript += transcription.lower() + " "
|
42 |
+
|
43 |
+
return transcript
|
44 |
+
|
45 |
+
gradio_ui = gr.Interface(
|
46 |
+
fn=asr_transcript,
|
47 |
+
title="Speech-to-Text with HuggingFace+Wav2Vec2",
|
48 |
+
description="Upload an audio clip in Russian, English, or French and let AI do the hard work of transcribing",
|
49 |
+
inputs = [gr.inputs.Audio(label="Upload Audio File", type="file"),
|
50 |
+
gr.inputs.Radio(label="Pick an STT Model - (language)",
|
51 |
+
choices=["English",
|
52 |
+
"Russian",
|
53 |
+
"French"])],
|
54 |
+
outputs=gr.outputs.Textbox(label="Auto-Transcript"),
|
55 |
+
)
|
56 |
+
|
57 |
+
gradio_ui.launch()
|