File size: 5,943 Bytes
cd7aa15
fcd8965
cd7aa15
 
 
 
49df20f
723513d
cd7aa15
 
 
 
 
49df20f
cd7aa15
f6dc6c7
f0a5b40
cd7aa15
8d55ac9
723513d
49df20f
f6dc6c7
551152e
 
f6dc6c7
551152e
723513d
 
 
cd7aa15
 
 
 
f0a5b40
 
 
f6dc6c7
098a61e
cd7aa15
3a9d859
cd7aa15
3a79217
 
098a61e
cd7aa15
f6dc6c7
cd7aa15
 
f6dc6c7
 
 
f0a5b40
771c2e9
1bb8243
771c2e9
f6dc6c7
 
 
 
 
 
1bb8243
771c2e9
 
f6dc6c7
 
 
 
 
771c2e9
3a9d859
 
 
 
 
 
941924a
3a9d859
 
 
 
f6dc6c7
 
3a9d859
 
 
f6dc6c7
 
3a9d859
f6dc6c7
3a9d859
 
f6dc6c7
3a9d859
f6dc6c7
 
 
 
 
 
 
 
 
 
 
 
 
941924a
f6dc6c7
 
 
3a9d859
f6dc6c7
f0a5b40
cd7aa15
f6dc6c7
cd7aa15
f6dc6c7
 
 
 
f0a5b40
cd7aa15
f6dc6c7
cd7aa15
f6dc6c7
f0a5b40
 
 
 
cd7aa15
 
f0a5b40
 
cd7aa15
f0a5b40
f6dc6c7
 
 
 
 
 
 
 
a312467
f6dc6c7
49df20f
f6dc6c7
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
import os
import tarfile
import torch
import torchaudio
import numpy as np
import streamlit as st
import matplotlib.pyplot as plt
from huggingface_hub import login
from transformers import (
    AutoProcessor,
    AutoModelForSpeechSeq2Seq,
    TrainingArguments,
    Trainer,
    DataCollatorForSeq2Seq,
)
from cryptography.fernet import Fernet

# ================================
# 1️⃣ Authenticate with Hugging Face Hub (Securely)
# ================================
HF_TOKEN = os.getenv("hf_token")  

if HF_TOKEN is None:
    raise ValueError("❌ Hugging Face API token not found. Please set it in Secrets.")

login(token=HF_TOKEN)

# ================================
# 2️⃣ Load Model & Processor
# ================================
MODEL_NAME = "AqeelShafy7/AudioSangraha-Audio_to_Text"
processor = AutoProcessor.from_pretrained(MODEL_NAME)
model = AutoModelForSpeechSeq2Seq.from_pretrained(MODEL_NAME)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
print(f"βœ… Model loaded on {device}")

# ================================
# 3️⃣ Load Dataset (From Extracted Folder)
# ================================
DATASET_TAR_PATH = "dev-clean.tar.gz"
EXTRACT_PATH = "./librispeech_dev_clean"

if not os.path.exists(EXTRACT_PATH):
    print("πŸ”„ Extracting dataset...")
    with tarfile.open(DATASET_TAR_PATH, "r:gz") as tar:
        tar.extractall(EXTRACT_PATH)
    print("βœ… Extraction complete.")
else:
    print("βœ… Dataset already extracted.")

AUDIO_FOLDER = os.path.join(EXTRACT_PATH, "LibriSpeech", "dev-clean")

def find_audio_files(base_folder):
    audio_files = []
    for root, _, files in os.walk(base_folder):
        for file in files:
            if file.endswith(".flac"):
                audio_files.append(os.path.join(root, file))
    return audio_files

audio_files = find_audio_files(AUDIO_FOLDER)

if not audio_files:
    raise FileNotFoundError(f"❌ No .flac files found in {AUDIO_FOLDER}. Check dataset structure!")

print(f"βœ… Found {len(audio_files)} audio files in dataset!")

# ================================
# 4️⃣ Load Transcripts
# ================================
def load_transcripts():
    transcript_dict = {}
    for root, _, files in os.walk(AUDIO_FOLDER):
        for file in files:
            if file.endswith(".txt"):
                with open(os.path.join(root, file), "r", encoding="utf-8") as f:
                    for line in f:
                        parts = line.strip().split(" ", 1)
                        if len(parts) == 2:
                            file_id, text = parts
                            transcript_dict[file_id] = text
    return transcript_dict

transcripts = load_transcripts()
if not transcripts:
    raise FileNotFoundError("❌ No transcripts found! Check dataset structure.")

print(f"βœ… Loaded {len(transcripts)} transcripts.")

# ================================
# 5️⃣ Preprocess Dataset (Fixing `input_ids` issue)
# ================================
def load_and_process_audio(audio_path):
    waveform, sample_rate = torchaudio.load(audio_path)
    waveform = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(waveform)
    input_features = processor(waveform.squeeze().numpy(), sampling_rate=16000, return_tensors="pt").input_features[0]
    return input_features

dataset = []
for audio_file in audio_files[:100]:
    file_id = os.path.basename(audio_file).replace(".flac", "")
    if file_id in transcripts:
        input_features = load_and_process_audio(audio_file)
        labels = processor.tokenizer(transcripts[file_id], padding="max_length", truncation=True, return_tensors="pt").input_ids[0]
        dataset.append({"input_features": input_features, "labels": labels})

train_size = int(0.8 * len(dataset))
train_dataset = dataset[:train_size]
eval_dataset = dataset[train_size:]

print(f"βœ… Dataset Prepared! Training: {len(train_dataset)}, Evaluation: {len(eval_dataset)}")

# ================================
# 6️⃣ Streamlit UI: Fine-Tuning Hyperparameter Selection
# ================================
st.sidebar.title("πŸ”§ Fine-Tuning Hyperparameters")
num_epochs = st.sidebar.slider("Epochs", min_value=1, max_value=10, value=3)
learning_rate = st.sidebar.select_slider("Learning Rate", options=[5e-4, 1e-4, 5e-5, 1e-5], value=5e-5)
batch_size = st.sidebar.select_slider("Batch Size", options=[2, 4, 8, 16], value=8)

# ================================
# 7️⃣ Streamlit ASR Web App (Fast Decoding & Adversarial Attack Detection)
# ================================
st.title("πŸŽ™οΈ Speech-to-Text ASR Model with Security Features 🎢")

audio_file = st.file_uploader("Upload an audio file", type=["wav", "mp3", "flac"])

if audio_file:
    audio_path = "temp_audio.wav"
    with open(audio_path, "wb") as f:
        f.write(audio_file.read())

    waveform, sample_rate = torchaudio.load(audio_path)
    waveform = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(waveform)
    
    # Simulate an adversarial attack by injecting random noise
    attack_strength = st.sidebar.slider("Attack Strength", 0.0, 0.1, 0.2, 0.5, 0.7,0.9)
    adversarial_waveform = waveform + (attack_strength * torch.randn_like(waveform))
    adversarial_waveform = torch.clamp(adversarial_waveform, -1.0, 1.0)
    
    input_features = processor(adversarial_waveform.squeeze().numpy(), sampling_rate=16000, return_tensors="pt").input_features.to(device)
    
    with torch.inference_mode():
        generated_ids = model.generate(input_features, max_length=200, num_beams=2, do_sample=False, use_cache=True, attention_mask=torch.ones(input_features.shape, dtype=torch.long).to(device))
        transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    
    if attack_strength > 0.1:
        st.warning("⚠️ Adversarial attack detected! Transcription secured.")
    
    st.success("πŸ“„ Secure Transcription:")
    st.write(transcription)