Update src/streamlit_app.py
Browse files- src/streamlit_app.py +46 -97
src/streamlit_app.py
CHANGED
@@ -1,165 +1,114 @@
|
|
1 |
import streamlit as st
|
2 |
import os
|
3 |
-
import numpy as np # linear algebra
|
4 |
-
import pandas as pd # data processing
|
5 |
# from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2Processor
|
6 |
# from utils import download_video, extract_audio, accent_classify
|
7 |
-
import whisper
|
8 |
from transformers import pipeline
|
|
|
|
|
|
|
9 |
import yt_dlp
|
10 |
import torchaudio
|
11 |
-
import yt_dlp
|
12 |
import ffmpeg
|
13 |
-
from transformers.utils import logging
|
14 |
|
15 |
logging.set_verbosity_info()
|
16 |
|
17 |
-
# Define the resampling rate in Hertz (Hz) for audio data
|
18 |
RATE_HZ = 16000
|
19 |
-
# Define the maximum audio interval length to consider in seconds
|
20 |
MAX_SECONDS = 1
|
21 |
-
# Calculate the maximum audio interval length in samples by multiplying the rate and seconds
|
22 |
MAX_LENGTH = RATE_HZ * MAX_SECONDS
|
23 |
|
24 |
|
25 |
-
def download_video(url,
|
26 |
-
os.makedirs(output_dir, exist_ok=True)
|
27 |
ydl_opts = {
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
}
|
37 |
|
38 |
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
39 |
ydl.download([url])
|
40 |
-
return
|
41 |
|
42 |
-
def extract_audio(input_path,
|
43 |
-
os.makedirs(output_dir, exist_ok=True)
|
44 |
-
output_path = os.path.join(output_dir, "audio.mp3")
|
45 |
(
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
return output_path
|
53 |
|
54 |
-
# Split files by chunks with == MAX_LENGTH size
|
55 |
def split_audio(file):
|
56 |
try:
|
57 |
-
# Load the audio file using torchaudio and get its sample rate.
|
58 |
audio, rate = torchaudio.load(str(file))
|
59 |
-
|
60 |
-
# Calculate the number of segments based on the MAX_LENGTH
|
61 |
num_segments = (len(audio[0]) // MAX_LENGTH) # Floor division to get segments
|
62 |
-
|
63 |
-
# Create an empty list to store segmented audio data
|
64 |
segmented_audio = []
|
65 |
-
|
66 |
-
# Split the audio into segments
|
67 |
for i in range(num_segments):
|
68 |
start = i * MAX_LENGTH
|
69 |
end = min((i + 1) * MAX_LENGTH, len(audio[0]))
|
70 |
segment = audio[0][start:end]
|
71 |
-
|
72 |
-
# Create a transformation to resample the audio to a specified sample rate (RATE_HZ).
|
73 |
transform = torchaudio.transforms.Resample(rate, RATE_HZ)
|
74 |
segment = transform(segment).squeeze(0).numpy().reshape(-1)
|
75 |
-
|
76 |
segmented_audio.append(segment)
|
77 |
-
|
78 |
-
# Create a DataFrame from the segmented audio
|
79 |
df_segments = pd.DataFrame({'audio': segmented_audio})
|
80 |
-
|
81 |
return df_segments
|
82 |
|
83 |
except Exception as e:
|
84 |
-
# If an exception occurs (e.g., file not found), return nothing
|
85 |
print(f"Error processing file: {e}")
|
86 |
return None
|
87 |
|
88 |
def accent_classify(pipe, audio_path):
|
89 |
audio_df = split_audio(audio_path)
|
90 |
-
return pipe(np.concatenate(audio_df["audio"][:
|
91 |
|
92 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
|
|
|
94 |
st.title("🎙️ English Accent Classifier")
|
95 |
st.markdown("Upload a video link and get the English accent with confidence.")
|
96 |
-
|
97 |
-
st.subheader("1. Upload a Video File")
|
98 |
-
uploaded_file = st.file_uploader("Choose a video file", type=["mp4", "mov", "avi"])
|
99 |
-
|
100 |
-
st.subheader("2. Or Enter a Video URL")
|
101 |
video_url = st.text_input("Paste a public video URL (Loom, or MP4):")
|
102 |
|
103 |
if st.button("Analyze"):
|
104 |
-
|
105 |
-
|
106 |
-
os.makedirs(output_dir, exist_ok=True)
|
107 |
-
|
108 |
-
if uploaded_file:
|
109 |
-
video_path = os.path.join(output_dir, "video.mp4")
|
110 |
-
with open(video_path, "wb") as f:
|
111 |
-
f.write(uploaded_file.read())
|
112 |
-
st.success("✅ Video uploaded successfully.")
|
113 |
-
elif video_url.strip():
|
114 |
-
with st.spinner("Downloading video from URL..."):
|
115 |
-
try:
|
116 |
-
video_path = download_video(video_url)
|
117 |
-
except Exception as e:
|
118 |
-
st.error(f"❌ Failed to download video: {e}")
|
119 |
-
else:
|
120 |
-
st.success(f"✅ Video downloaded: {video_path}")
|
121 |
-
|
122 |
else:
|
123 |
-
st.
|
|
|
124 |
|
125 |
-
if video_path and os.path.exists(video_path):
|
126 |
-
st.write("Exists:", os.path.exists(video_path))
|
127 |
with st.spinner("Extracting audio..."):
|
128 |
audio_path = extract_audio(video_path)
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
# pass
|
138 |
-
|
139 |
-
with st.spinner("Extracting waves..."):
|
140 |
-
audio_df = split_audio(audio_path)
|
141 |
-
# print(np.concatenate(audio_df["audio"][:50].to_list()))
|
142 |
-
waves = f"{np.concatenate(audio_df["audio"][:5].to_list())}"
|
143 |
-
st.markdown("**Audio waves:**")
|
144 |
-
st.text_area("Audio waves", waves, height=200)
|
145 |
-
|
146 |
-
|
147 |
-
with st.spinner("Classifying accent..."):
|
148 |
model_name = "dima806/english_accents_classification"
|
149 |
pipe = pipeline('audio-classification', model=model_name, device=0)
|
150 |
accent_data = accent_classify(pipe, audio_path)
|
151 |
-
|
152 |
-
|
153 |
-
accent = accent_data.get("label", "American")
|
154 |
-
confidence = accent_data.get("score", 0.0)
|
155 |
-
# pass
|
156 |
|
157 |
st.success("Analysis Complete!")
|
158 |
st.markdown(f"**Accent:** {accent}")
|
159 |
st.markdown(f"**Confidence Score:** {confidence:.2f}%")
|
|
|
160 |
# st.markdown("**Transcription:**")
|
161 |
# st.text_area("Transcript", transcription, height=200)
|
162 |
|
163 |
# Cleanup
|
164 |
os.remove(video_path)
|
165 |
-
os.remove(audio_path)
|
|
|
1 |
import streamlit as st
|
2 |
import os
|
|
|
|
|
3 |
# from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2Processor
|
4 |
# from utils import download_video, extract_audio, accent_classify
|
5 |
+
# import whisper
|
6 |
from transformers import pipeline
|
7 |
+
from transformers.utils import logging
|
8 |
+
import numpy as np
|
9 |
+
import pandas as pd
|
10 |
import yt_dlp
|
11 |
import torchaudio
|
|
|
12 |
import ffmpeg
|
|
|
13 |
|
14 |
logging.set_verbosity_info()
|
15 |
|
|
|
16 |
RATE_HZ = 16000
|
|
|
17 |
MAX_SECONDS = 1
|
|
|
18 |
MAX_LENGTH = RATE_HZ * MAX_SECONDS
|
19 |
|
20 |
|
21 |
+
def download_video(url, output_path="video.mp4"):
|
|
|
22 |
ydl_opts = {
|
23 |
+
'format': 'worstvideo[ext=mp4]+bestaudio[ext=m4a]/bestaudio',
|
24 |
+
'outtmpl': output_path,
|
25 |
+
'merge_output_format': 'mp4',
|
26 |
+
'quiet': True,
|
27 |
+
'noplaylist': True,
|
28 |
+
'nocheckcertificate': True,
|
29 |
+
'retries': 3,
|
30 |
+
}
|
|
|
31 |
|
32 |
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
33 |
ydl.download([url])
|
34 |
+
return output_path
|
35 |
|
36 |
+
def extract_audio(input_path, output_path="audio.mp3"):
|
|
|
|
|
37 |
(
|
38 |
+
ffmpeg
|
39 |
+
.input(input_path)
|
40 |
+
.output(output_path, format='mp3', acodec='libmp3lame', audio_bitrate='192k')
|
41 |
+
.overwrite_output()
|
42 |
+
.run(quiet=True)
|
43 |
+
)
|
44 |
return output_path
|
45 |
|
|
|
46 |
def split_audio(file):
|
47 |
try:
|
|
|
48 |
audio, rate = torchaudio.load(str(file))
|
|
|
|
|
49 |
num_segments = (len(audio[0]) // MAX_LENGTH) # Floor division to get segments
|
|
|
|
|
50 |
segmented_audio = []
|
|
|
|
|
51 |
for i in range(num_segments):
|
52 |
start = i * MAX_LENGTH
|
53 |
end = min((i + 1) * MAX_LENGTH, len(audio[0]))
|
54 |
segment = audio[0][start:end]
|
|
|
|
|
55 |
transform = torchaudio.transforms.Resample(rate, RATE_HZ)
|
56 |
segment = transform(segment).squeeze(0).numpy().reshape(-1)
|
|
|
57 |
segmented_audio.append(segment)
|
|
|
|
|
58 |
df_segments = pd.DataFrame({'audio': segmented_audio})
|
|
|
59 |
return df_segments
|
60 |
|
61 |
except Exception as e:
|
|
|
62 |
print(f"Error processing file: {e}")
|
63 |
return None
|
64 |
|
65 |
def accent_classify(pipe, audio_path):
|
66 |
audio_df = split_audio(audio_path)
|
67 |
+
return pipe(np.concatenate(audio_df["audio"][:250].to_list()))[0]
|
68 |
|
69 |
+
accent_mapping = {
|
70 |
+
'us': 'American',
|
71 |
+
'canada': 'Canadian',
|
72 |
+
'england': 'British',
|
73 |
+
'indian': 'Indian',
|
74 |
+
'australia': 'Australian',
|
75 |
+
}
|
76 |
|
77 |
+
st.set_page_config(page_title="Accent Classifier", layout="centered")
|
78 |
st.title("🎙️ English Accent Classifier")
|
79 |
st.markdown("Upload a video link and get the English accent with confidence.")
|
|
|
|
|
|
|
|
|
|
|
80 |
video_url = st.text_input("Paste a public video URL (Loom, or MP4):")
|
81 |
|
82 |
if st.button("Analyze"):
|
83 |
+
if not video_url.strip():
|
84 |
+
st.warning("Please enter a valid URL.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
else:
|
86 |
+
with st.spinner("Downloading video..."):
|
87 |
+
video_path = download_video(video_url)
|
88 |
|
|
|
|
|
89 |
with st.spinner("Extracting audio..."):
|
90 |
audio_path = extract_audio(video_path)
|
91 |
+
|
92 |
+
# with st.spinner("Transcribing with Whisper..."):
|
93 |
+
# whisper_model = whisper.load_model("base")
|
94 |
+
# result = whisper_model.transcribe(audio_path)
|
95 |
+
# transcription = result['text']
|
96 |
+
# # pass
|
97 |
+
|
98 |
+
with st.spinner("Classifying accent..."):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
model_name = "dima806/english_accents_classification"
|
100 |
pipe = pipeline('audio-classification', model=model_name, device=0)
|
101 |
accent_data = accent_classify(pipe, audio_path)
|
102 |
+
accent = accent_mapping.get(accent_data.get("label", "us"))
|
103 |
+
confidence = accent_data.get("score", 0)
|
|
|
|
|
|
|
104 |
|
105 |
st.success("Analysis Complete!")
|
106 |
st.markdown(f"**Accent:** {accent}")
|
107 |
st.markdown(f"**Confidence Score:** {confidence:.2f}%")
|
108 |
+
|
109 |
# st.markdown("**Transcription:**")
|
110 |
# st.text_area("Transcript", transcription, height=200)
|
111 |
|
112 |
# Cleanup
|
113 |
os.remove(video_path)
|
114 |
+
os.remove(audio_path)
|