changed speechlib to whisperx
Browse files
app.py
CHANGED
@@ -3,43 +3,52 @@ import base64
|
|
3 |
import os
|
4 |
import json
|
5 |
import streamlit as st
|
6 |
-
|
|
|
7 |
|
8 |
-
def
|
9 |
-
transcriptor = Transcriptor(file, log_folder, language, modelSize, ACCESS_TOKEN, voices_folder, quantization)
|
10 |
-
return transcriptor.whisper()
|
11 |
-
|
12 |
-
def transform_transcript(transcript):
|
13 |
result = []
|
14 |
-
for segment in
|
15 |
-
|
16 |
-
|
|
|
|
|
|
|
|
|
17 |
return '\n'.join(result)
|
18 |
|
19 |
st.title('Audio Transcription App')
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
ACCESS_TOKEN = st.secrets["HF_TOKEN"]
|
22 |
|
23 |
uploaded_file = st.file_uploader("Загрузите аудиофайл", type=["mp4", "wav", "m4a"])
|
24 |
|
25 |
if uploaded_file is not None:
|
|
|
26 |
file_extension = uploaded_file.name.split(".")[-1] # Получаем расширение файла
|
27 |
temp_file_path = f"temp_file.{file_extension}" # Создаем временное имя файла с правильным расширением
|
28 |
|
29 |
with open(temp_file_path, "wb") as f:
|
30 |
f.write(uploaded_file.getbuffer())
|
31 |
-
|
32 |
-
log_folder = "logs"
|
33 |
-
language = "ru"
|
34 |
-
modelSize = os.getenv('WHISPER_MODEL_SIZE')
|
35 |
-
voices_folder = ""
|
36 |
-
quantization = False
|
37 |
|
38 |
with st.spinner('Транскрибируем...'):
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
st.write("Результат транскрибации:")
|
42 |
-
transcript =
|
43 |
st.text(transcript)
|
44 |
|
45 |
with st.spinner('Резюмируем...'):
|
|
|
3 |
import os
|
4 |
import json
|
5 |
import streamlit as st
|
6 |
+
import whisperx
|
7 |
+
import torch
|
8 |
|
9 |
+
def convert_segments_to_text(segments):
|
|
|
|
|
|
|
|
|
10 |
result = []
|
11 |
+
for segment in segments:
|
12 |
+
speaker = segment['speaker']
|
13 |
+
start = segment['start']
|
14 |
+
end = segment['end']
|
15 |
+
text = segment['text']
|
16 |
+
formatted_text = f'{speaker} ({start} : {end}) : {text}'
|
17 |
+
result.append(formatted_text)
|
18 |
return '\n'.join(result)
|
19 |
|
20 |
st.title('Audio Transcription App')
|
21 |
+
st.sidebar.title("Settings")
|
22 |
+
# Sidebar inputs
|
23 |
+
device = st.sidebar.selectbox("Device", ["cpu", "cuda"], index=1)
|
24 |
+
batch_size = st.sidebar.number_input("Batch Size", min_value=1, value=16)
|
25 |
+
compute_type = st.sidebar.selectbox("Compute Type", ["float16", "int8"], index=0)
|
26 |
|
27 |
ACCESS_TOKEN = st.secrets["HF_TOKEN"]
|
28 |
|
29 |
uploaded_file = st.file_uploader("Загрузите аудиофайл", type=["mp4", "wav", "m4a"])
|
30 |
|
31 |
if uploaded_file is not None:
|
32 |
+
st.audio(uploaded_file)
|
33 |
file_extension = uploaded_file.name.split(".")[-1] # Получаем расширение файла
|
34 |
temp_file_path = f"temp_file.{file_extension}" # Создаем временное имя файла с правильным расширением
|
35 |
|
36 |
with open(temp_file_path, "wb") as f:
|
37 |
f.write(uploaded_file.getbuffer())
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
with st.spinner('Транскрибируем...'):
|
40 |
+
# Load model
|
41 |
+
model = whisperx.load_model("medium", device, compute_type=compute_type)
|
42 |
+
# Load and transcribe audio
|
43 |
+
audio = whisperx.load_audio(temp_file_path)
|
44 |
+
result = model.transcribe(audio, batch_size=batch_size, language="Russian")
|
45 |
+
# Load diarization model (replace YOUR_HF_TOKEN with actual token)
|
46 |
+
diarize_model = whisperx.DiarizationPipeline(use_auth_token=st.secrets["HF_TOKEN"], device=device)
|
47 |
+
diarize_segments = diarize_model(audio)
|
48 |
+
result = whisperx.assign_word_speakers(diarize_segments, result)
|
49 |
|
50 |
st.write("Результат транскрибации:")
|
51 |
+
transcript = convert_segments_to_text(result)
|
52 |
st.text(transcript)
|
53 |
|
54 |
with st.spinner('Резюмируем...'):
|