transcribe with whisperx
Browse files- transcribe.py +44 -12
transcribe.py
CHANGED
@@ -1,20 +1,46 @@
|
|
1 |
import os
|
2 |
import argparse
|
|
|
3 |
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
output_folder = "transcriptions"
|
6 |
|
7 |
# Transcribe audio file
|
8 |
model = "large-v2"
|
9 |
-
word_timestamps = True
|
|
|
|
|
10 |
fp16 = False
|
11 |
-
|
12 |
verbose = False
|
|
|
|
|
13 |
threads = 4
|
14 |
output_format = "srt"
|
15 |
-
|
16 |
-
|
17 |
-
--
|
|
|
|
|
|
|
18 |
os.system(command)
|
19 |
|
20 |
if __name__ == "__main__":
|
@@ -22,10 +48,10 @@ if __name__ == "__main__":
|
|
22 |
parser.add_argument('input_files', help='Input audio files')
|
23 |
parser.add_argument('language', help='Language of the audio file')
|
24 |
parser.add_argument('speakers_file', help='File with the number of speakers')
|
|
|
25 |
args = parser.parse_args()
|
26 |
|
27 |
vocals_folder = "vocals"
|
28 |
-
extension = "wav"
|
29 |
|
30 |
with open(args.speakers_file, 'r') as f:
|
31 |
speakers = f.read().splitlines()
|
@@ -34,8 +60,14 @@ if __name__ == "__main__":
|
|
34 |
with open(args.input_files, 'r') as f:
|
35 |
inputs = f.read().splitlines()
|
36 |
for input in inputs:
|
37 |
-
|
38 |
-
_, input_name =
|
39 |
-
|
40 |
-
|
41 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
import argparse
|
3 |
+
from lang_list import LANGUAGE_NAME_TO_CODE, WHISPER_LANGUAGES
|
4 |
|
5 |
+
# For pyannote.audio diarize
|
6 |
+
from pyannote.audio import Model
|
7 |
+
model = Model.from_pretrained("pyannote/segmentation-3.0", use_auth_token="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")
|
8 |
+
|
9 |
+
language_dict = {}
|
10 |
+
# Iterate over the LANGUAGE_NAME_TO_CODE dictionary
|
11 |
+
for language_name, language_code in LANGUAGE_NAME_TO_CODE.items():
|
12 |
+
# Extract the language code (the first two characters before the underscore)
|
13 |
+
lang_code = language_code.split('_')[0].lower()
|
14 |
+
|
15 |
+
# Check if the language code is present in WHISPER_LANGUAGES
|
16 |
+
if lang_code in WHISPER_LANGUAGES:
|
17 |
+
# Construct the entry for the resulting dictionary
|
18 |
+
language_dict[language_name] = {
|
19 |
+
"transcriber": lang_code,
|
20 |
+
"translator": language_code
|
21 |
+
}
|
22 |
+
|
23 |
+
def transcribe(audio_file, language, device):
|
24 |
output_folder = "transcriptions"
|
25 |
|
26 |
# Transcribe audio file
|
27 |
model = "large-v2"
|
28 |
+
# word_timestamps = True
|
29 |
+
print_progress = True
|
30 |
+
compute_type = "float32"
|
31 |
fp16 = False
|
32 |
+
batch_size = 8
|
33 |
verbose = False
|
34 |
+
min_speakers = 1
|
35 |
+
max_speakers = 10
|
36 |
threads = 4
|
37 |
output_format = "srt"
|
38 |
+
hf_token = "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"
|
39 |
+
command = f'whisperx {audio_file} --model {model} --batch_size {batch_size} --compute_type {compute_type} \
|
40 |
+
--output_dir {output_folder} --output_format {output_format} --verbose {verbose} --language {language} \
|
41 |
+
--fp16 {fp16} --threads {threads} --print_progress {print_progress} --min_speakers {min_speakers} \
|
42 |
+
--max_speakers {max_speakers} --diarize --hf_token {hf_token}'
|
43 |
+
# --diarize'
|
44 |
os.system(command)
|
45 |
|
46 |
if __name__ == "__main__":
|
|
|
48 |
parser.add_argument('input_files', help='Input audio files')
|
49 |
parser.add_argument('language', help='Language of the audio file')
|
50 |
parser.add_argument('speakers_file', help='File with the number of speakers')
|
51 |
+
parser.add_argument('device', help='Device to use for PyTorch inference')
|
52 |
args = parser.parse_args()
|
53 |
|
54 |
vocals_folder = "vocals"
|
|
|
55 |
|
56 |
with open(args.speakers_file, 'r') as f:
|
57 |
speakers = f.read().splitlines()
|
|
|
60 |
with open(args.input_files, 'r') as f:
|
61 |
inputs = f.read().splitlines()
|
62 |
for input in inputs:
|
63 |
+
input_file, _ = input.split('.')
|
64 |
+
_, input_name = input_file.split('/')
|
65 |
+
if speakers > 0:
|
66 |
+
extension = "wav"
|
67 |
+
for i in range(speakers):
|
68 |
+
file = f'{vocals_folder}/{input_name}_speaker{i:003d}.{extension}'
|
69 |
+
transcribe(file, language_dict[args.language]["transcriber"], args.device)
|
70 |
+
else:
|
71 |
+
extension = "mp3"
|
72 |
+
file = f'{vocals_folder}/{input_name}.{extension}'
|
73 |
+
transcribe(file, language_dict[args.language]["transcriber"], args.device)
|