Spaces:
Build error
Build error
patrickvonplaten
commited on
Commit
•
00349e4
1
Parent(s):
5b8b578
Update app.py
Browse files
app.py
CHANGED
@@ -2,9 +2,11 @@ import gradio as gr
|
|
2 |
import librosa
|
3 |
from transformers import AutoFeatureExtractor, AutoTokenizer, SpeechEncoderDecoderModel
|
4 |
|
5 |
-
feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15")
|
6 |
-
tokenizer = AutoTokenizer.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_fast=False)
|
7 |
-
model = SpeechEncoderDecoderModel.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15")
|
|
|
|
|
8 |
|
9 |
def process_audio_file(file):
|
10 |
data, sr = librosa.load(file)
|
@@ -14,11 +16,18 @@ def process_audio_file(file):
|
|
14 |
input_values = feature_extractor(data, return_tensors="pt").input_values
|
15 |
return input_values
|
16 |
|
17 |
-
def transcribe(
|
18 |
|
19 |
target_code = target_language.split("(")[-1].split(")")[0]
|
20 |
forced_bos_token_id = MAPPING[target_code]
|
21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
input_values = process_audio_file(file)
|
23 |
|
24 |
sequences = model.generate(input_values, forced_bos_token_id=forced_bos_token_id)
|
@@ -65,7 +74,8 @@ MAPPING = {
|
|
65 |
iface = gr.Interface(
|
66 |
fn=transcribe,
|
67 |
inputs=[
|
68 |
-
gr.inputs.Audio(source="microphone", type='
|
|
|
69 |
gr.inputs.Dropdown(target_language),
|
70 |
],
|
71 |
outputs="text",
|
|
|
2 |
import librosa
|
3 |
from transformers import AutoFeatureExtractor, AutoTokenizer, SpeechEncoderDecoderModel
|
4 |
|
5 |
+
# feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15")
|
6 |
+
# tokenizer = AutoTokenizer.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_fast=False)
|
7 |
+
# model = SpeechEncoderDecoderModel.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15")
|
8 |
+
|
9 |
+
feature_extractor = tokenizer = model = None
|
10 |
|
11 |
def process_audio_file(file):
|
12 |
data, sr = librosa.load(file)
|
|
|
16 |
input_values = feature_extractor(data, return_tensors="pt").input_values
|
17 |
return input_values
|
18 |
|
19 |
+
def transcribe(file_mic, file_upload, target_language):
|
20 |
|
21 |
target_code = target_language.split("(")[-1].split(")")[0]
|
22 |
forced_bos_token_id = MAPPING[target_code]
|
23 |
+
|
24 |
+
if file_mic is not None and file_upload is not None:
|
25 |
+
print("Warning: You've uploaded an audio file and used the microphone. The recorded file from the microphone will be used and the uploaded audio will be discarded.")
|
26 |
+
elif file_mic is None and file_upload is None:
|
27 |
+
raise ValueError("You have to either use the microphone or upload an audio file")
|
28 |
+
|
29 |
+
file = file_mic or file_upload
|
30 |
+
|
31 |
input_values = process_audio_file(file)
|
32 |
|
33 |
sequences = model.generate(input_values, forced_bos_token_id=forced_bos_token_id)
|
|
|
74 |
iface = gr.Interface(
|
75 |
fn=transcribe,
|
76 |
inputs=[
|
77 |
+
gr.inputs.Audio(source="microphone", type='filepath_mic', optional=True),
|
78 |
+
gr.inputs.Audio(source="upload", type='filepath_upload', optional=True),
|
79 |
gr.inputs.Dropdown(target_language),
|
80 |
],
|
81 |
outputs="text",
|