File size: 23,724 Bytes
e1bae5b 4c0be85 e368f8b fb79caf e368f8b 8cc69ea 0427f41 8cc69ea fb79caf e368f8b fb79caf e368f8b e8ce33d e368f8b e8ce33d e368f8b e8ce33d 341746e e368f8b e8ce33d e368f8b e8ce33d e368f8b b815c4a e8ce33d e368f8b f427fe9 8cc69ea 1a539e2 e368f8b 0427f41 3fc92f2 0427f41 e368f8b 0427f41 e368f8b 0427f41 e368f8b 0427f41 2caaec7 6d2ca12 8cc69ea 1a539e2 f4644ed e368f8b 1a539e2 e368f8b 1a539e2 e368f8b 1a539e2 ef745e1 f890d9b ef745e1 e368f8b 6d2ca12 8cc69ea 1a539e2 e368f8b 0427f41 c1cd1f5 e368f8b ef745e1 e499054 9eda364 e368f8b 7f5deab 8cc69ea e368f8b 8cc69ea e368f8b 470c6fe e368f8b 470c6fe 7fc8342 f4644ed e368f8b 470c6fe e368f8b 9eda364 e368f8b 7f5deab e368f8b 375457e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 |
import os
import json
import gradio as gr
import torch
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
import logging
import traceback
import sys
from audio_processing import AudioProcessor
import spaces
from chunkedTranscriber import ChunkedTranscriber
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
handlers=[logging.StreamHandler(sys.stdout)]
)
logger = logging.getLogger(__name__)
def load_qa_model():
"""Load question-answering model with support for long input contexts."""
try:
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4"
# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=os.getenv("HF_TOKEN"))
tokenizer.model_max_length = 8192 # Ensure the tokenizer can handle 8192 tokens
# Load the model
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
device_map="auto",
rope_scaling={
"type": "dynamic", # Ensure compatibility with long contexts
"factor": 8.0
},
use_auth_token=os.getenv("HF_TOKEN")
)
# Load the pipeline
qa_pipeline = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
max_new_tokens=4096, # Adjust as needed for your use case
)
return qa_pipeline
except Exception as e:
logger.error(f"Failed to load Q&A model: {str(e)}")
return None
# def load_qa_model():
# """Load question-answering model"""
# try:
# model_id = "meta-llama/Meta-Llama-3-8B-Instruct"
# qa_pipeline = pipeline(
# "text-generation",
# model="hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
# model_kwargs={"torch_dtype": torch.bfloat16},
# device_map="auto",
# use_auth_token=os.getenv("HF_TOKEN")
# )
# return qa_pipeline
# except Exception as e:
# logger.error(f"Failed to load Q&A model: {str(e)}")
# return None
def load_summarization_model():
"""Load summarization model"""
try:
summarizer = pipeline(
"summarization",
model="sshleifer/distilbart-cnn-12-6",
device=0 if torch.cuda.is_available() else -1
)
return summarizer
except Exception as e:
logger.error(f"Failed to load summarization model: {str(e)}")
return None
@spaces.GPU(duration=120)
def process_audio(audio_file, translate=False):
"""Process audio file"""
transcriber = ChunkedTranscriber(chunk_size=5, overlap=1)
_translation, _output = transcriber.transcribe_audio(audio_file, translate=True)
return _translation, _output
# try:
# processor = AudioProcessor()
# language_segments, final_segments = processor.process_audio(audio_file, translate)
# # Format output
# transcription = ""
# full_text = ""
# # Add language detection information
# for segment in language_segments:
# transcription += f"Language: {segment['language']}\n"
# transcription += f"Time: {segment['start']:.2f}s - {segment['end']:.2f}s\n\n"
# # Add transcription/translation information
# transcription += "Transcription with language detection:\n\n"
# for segment in final_segments:
# transcription += f"[{segment['start']:.2f}s - {segment['end']:.2f}s] ({segment['language']}):\n"
# transcription += f"Original: {segment['text']}\n"
# if translate and 'translated' in segment:
# transcription += f"Translated: {segment['translated']}\n"
# full_text += segment['translated'] + " "
# else:
# full_text += segment['text'] + " "
# transcription += "\n"
# return transcription, full_text
# except Exception as e:
# logger.error(f"Audio processing failed: {str(e)}")
# raise gr.Error(f"Processing failed: {str(e)}")
@spaces.GPU(duration=120)
def summarize_text(text):
"""Summarize text"""
try:
summarizer = load_summarization_model()
if summarizer is None:
return "Summarization model could not be loaded."
logger.info("Successfully loaded summarization Model")
# logger.info(f"\n\n {text}\n")
summary = summarizer(text, max_length=150, min_length=50, do_sample=False)[0]['summary_text']
return summary
except Exception as e:
logger.error(f"Summarization failed: {str(e)}")
return "Error occurred during summarization."
@spaces.GPU(duration=120)
def answer_question(context, question):
"""Answer questions about the text"""
try:
qa_pipeline = load_qa_model()
if qa_pipeline is None:
return "Q&A model could not be loaded."
if not question :
return "Please enter your Question"
messages = [
# {"role": "system", "content": "You are a helpful assistant who can answer questions based on the given context."},
{"role":"system", "content": """
Analyze a translated transcript of a conversation that may contain multiple speakers and summarize the information in a structured intelligence document.
The input format will include word-level or sentence-level timestamps, each indicating the speaker ID, language, and translated text.
# Input Format Overview
Word-Level Timestamps Example:
```
[Start Time - End Time] - Speaker <ID> - Language: <Translated Language> - Translated Text: "<Word>"
```
Example:
```
0.01-0.02 - Speaker 1 - Language: English - Translated Text: "Proceed"
0.02-0.025 - Speaker 1 - Language: English - Translated Text: "with"
0.025-0.032 - Speaker 2 - Language: English - Translated Text: "caution"
```
Optional Sentence-Level Structure Example:
```
[Start Time - End Time] - Speaker <ID> - Language: <Translated Language> - Translated Text: "<Sentence>"
```
Example with Sentence Grouping:
```
0.01-0.05 - Speaker 1 - Language: English - Translated Text: "Proceed with caution."
0.06-0.12 - Speaker 2 - Language: English - Translated Text: "All systems are ready."
```
# Intelligence Summary Document Structure
Use the format below to create a structured summary for each conversation transcript received:
### 1. Top-Level Status & Assessment:
- **Threat Level Assessment**:
- Choose one:
- Completely Innocuous
- Likely Innocuous
- Unclear — Requires Investigation
- Likely Dangerous — Immediate Action
- Likely Dangerous — Delayed Action
- 100% Dangerous — Immediate Action
- 100% Dangerous — Delayed Action
- **Humanitarian Alert**: Identify any indications of distress, coercion, or need for assistance, such as signs of duress or requests for help.
### 2. Basic Metadata:
- **Number of Speakers**: Total and unique speakers detected.
- **Languages**: List of languages used, with indication of who spoke which language.
- **Location**: Actual or inferred locations of participants.
- **Communication Medium**: Identify the method of interaction (e.g., phone call, direct conversation).
### 3. Conversation Overview:
- **Summary**: Concise breakdown of the main points and context.
- **Alarming Keywords**: Identify any concerning words, including but not limited to keywords like "kill," "attack," "weapon," etc.
- **Suspicious or Cryptic Phrases**: Statements that appear coded or unclear in the context of the discussion.
### 4. In-Depth Analysis:
- **Network Connections**: Identify mentions of additional individuals or groups involved.
- **Intent & Emotional Tone Detection**: Analyze emotional cues (e.g., anger, fear, calmness, urgency). Identify signs of deception or tension.
- **Behavioral Patterns**: Highlight repeated themes, phrases, or signals of planning and coordination.
- **Code Words & Cryptic Language**: Detect terms that may indicate hidden or covert meaning.
- **Geolocation References**: Point out any inferences regarding regional language or place names.
- **Sentiment on Strategic Issues**: Identify any indication of radical, dissenting, or anti-national views that could imply unrest or extremism.
### 5. Resource Mentions & Operational Logistics:
- **Resource & Asset Mentions**: List any mention of tools, weapons, vehicles, or supply logistics.
- **Behavioral Deviations**: Identify shifts in tone, speech, or demeanor suggesting stress, coercion, urgency, or preparation.
### 6. Prioritization, Recommendations & Actionables:
- **High-Risk Alert Priority**: Identify whether the conversation should be flagged for further attention.
- **Recommended Actions**:
- **Surveillance**: Suggest surveillance if concerning patterns or keywords are detected.
- **Intervention**: Recommend intervention for urgent/high-risk cases.
- **Humanitarian Assistance**: Suggest immediate support for any signs of distress.
- **Follow-Up Analysis**: Identify statements that need deeper review for clarity or to understand potential hidden meanings.
# Steps
1. Analyze the input conversation for participant information and context.
2. Fill in each section of the Intelligence Summary Document structure.
3. Ensure all details, especially those related to potential risk factors or alerts, are captured and highlighted clearly.
# Output Format
Provide one structured Intelligence Summary Document for the conversation in either plain text format or structured JSON.
# JSON Format Example:
```json
{
"Top-Level Status & Assessment": {
"Threat Level Assessment": "Unclear - Requires Investigation",
"Humanitarian Alert": "No distress signals detected."
},
"Basic Metadata": {
"Number of Speakers": 2,
"Languages": {
"Speaker 1": "English",
"Speaker 2": "English"
},
"Location": "Unknown",
"Communication Medium": "Direct conversation"
},
"Conversation Overview": {
"Summary": "A cautious approach was suggested by Speaker 1, followed by an assurance from Speaker 2 that systems are ready.",
"Alarming Keywords": [],
"Suspicious or Cryptic Phrases": []
},
"In-Depth Analysis": {
"Network Connections": "None identified",
"Intent & Emotional Tone Detection": "Calm, precautionary tone",
"Behavioral Patterns": "Speaker 1 expressing concern, Speaker 2 providing assurance",
"Code Words & Cryptic Language": [],
"Geolocation References": [],
"Sentiment on Strategic Issues": "No radical or dissenting sentiment detected"
},
"Resource Mentions & Operational Logistics": {
"Resource & Asset Mentions": [],
"Behavioral Deviations": "None noted"
},
"Prioritization, Recommendations & Actionables": {
"High-Risk Alert Priority": "Low",
"Recommended Actions": {
"Surveillance": "No further surveillance needed.",
"Intervention": "Not required.",
"Humanitarian Assistance": "Not required.",
"Follow-Up Analysis": "No unusual phrases detected requiring review."
}
}
}
```
# Notes
- Ensure that you mark any ambiguous segments as requiring further investigation.
- Pay attention to emotional tone shifts or sudden changes in behavior.
- If any direct or implied threat is detected, prioritize appropriately using the provided classifications.
- Err on the side of caution. In case there is even a remote possibility that there might be something that required human attention, flag it.
Analyze a translated transcript of a conversation that may contain multiple speakers and summarize the information in a structured intelligence document.
The input format will include word-level or sentence-level timestamps, each indicating the speaker ID, language, and translated text.
# Input Format Overview
Word-Level Timestamps Example:
```
[Start Time - End Time] - Speaker <ID> - Language: <Translated Language> - Translated Text: "<Word>"
```
Example:
```
0.01-0.02 - Speaker 1 - Language: English - Translated Text: "Proceed"
0.02-0.025 - Speaker 1 - Language: English - Translated Text: "with"
0.025-0.032 - Speaker 2 - Language: English - Translated Text: "caution"
```
Optional Sentence-Level Structure Example:
```
[Start Time - End Time] - Speaker <ID> - Language: <Translated Language> - Translated Text: "<Sentence>"
```
Example with Sentence Grouping:
```
0.01-0.05 - Speaker 1 - Language: English - Translated Text: "Proceed with caution."
0.06-0.12 - Speaker 2 - Language: English - Translated Text: "All systems are ready."
```
# Intelligence Summary Document Structure
Use the format below to create a structured summary for each conversation transcript received:
### 1. Top-Level Status & Assessment:
- **Threat Level Assessment**:
- Choose one:
- Completely Innocuous
- Likely Innocuous
- Unclear — Requires Investigation
- Likely Dangerous — Immediate Action
- Likely Dangerous — Delayed Action
- 100% Dangerous — Immediate Action
- 100% Dangerous — Delayed Action
- **Humanitarian Alert**: Identify any indications of distress, coercion, or need for assistance, such as signs of duress or requests for help.
### 2. Basic Metadata:
- **Number of Speakers**: Total and unique speakers detected.
- **Languages**: List of languages used, with indication of who spoke which language.
- **Location**: Actual or inferred locations of participants.
- **Communication Medium**: Identify the method of interaction (e.g., phone call, direct conversation).
### 3. Conversation Overview:
- **Summary**: Concise breakdown of the main points and context.
- **Alarming Keywords**: Identify any concerning words, including but not limited to keywords like "kill," "attack," "weapon," etc.
- **Suspicious or Cryptic Phrases**: Statements that appear coded or unclear in the context of the discussion.
### 4. In-Depth Analysis:
- **Network Connections**: Identify mentions of additional individuals or groups involved.
- **Intent & Emotional Tone Detection**: Analyze emotional cues (e.g., anger, fear, calmness, urgency). Identify signs of deception or tension.
- **Behavioral Patterns**: Highlight repeated themes, phrases, or signals of planning and coordination.
- **Code Words & Cryptic Language**: Detect terms that may indicate hidden or covert meaning.
- **Geolocation References**: Point out any inferences regarding regional language or place names.
- **Sentiment on Strategic Issues**: Identify any indication of radical, dissenting, or anti-national views that could imply unrest or extremism.
### 5. Resource Mentions & Operational Logistics:
- **Resource & Asset Mentions**: List any mention of tools, weapons, vehicles, or supply logistics.
- **Behavioral Deviations**: Identify shifts in tone, speech, or demeanor suggesting stress, coercion, urgency, or preparation.
### 6. Prioritization, Recommendations & Actionables:
- **High-Risk Alert Priority**: Identify whether the conversation should be flagged for further attention.
- **Recommended Actions**:
- **Surveillance**: Suggest surveillance if concerning patterns or keywords are detected.
- **Intervention**: Recommend intervention for urgent/high-risk cases.
- **Humanitarian Assistance**: Suggest immediate support for any signs of distress.
- **Follow-Up Analysis**: Identify statements that need deeper review for clarity or to understand potential hidden meanings.
# Steps
1. Analyze the input conversation for participant information and context.
2. Fill in each section of the Intelligence Summary Document structure.
3. Ensure all details, especially those related to potential risk factors or alerts, are captured and highlighted clearly.
# Output Format
Provide one structured Intelligence Summary Document for the conversation in either plain text format or structured JSON.
# JSON Format Example:
```json
{
"Top-Level Status & Assessment": {
"Threat Level Assessment": "Unclear - Requires Investigation",
"Humanitarian Alert": "No distress signals detected."
},
"Basic Metadata": {
"Number of Speakers": 2,
"Languages": {
"Speaker 1": "English",
"Speaker 2": "English"
},
"Location": "Unknown",
"Communication Medium": "Direct conversation"
},
"Conversation Overview": {
"Summary": "A cautious approach was suggested by Speaker 1, followed by an assurance from Speaker 2 that systems are ready.",
"Alarming Keywords": [],
"Suspicious or Cryptic Phrases": []
},
"In-Depth Analysis": {
"Network Connections": "None identified",
"Intent & Emotional Tone Detection": "Calm, precautionary tone",
"Behavioral Patterns": "Speaker 1 expressing concern, Speaker 2 providing assurance",
"Code Words & Cryptic Language": [],
"Geolocation References": [],
"Sentiment on Strategic Issues": "No radical or dissenting sentiment detected"
},
"Resource Mentions & Operational Logistics": {
"Resource & Asset Mentions": [],
"Behavioral Deviations": "None noted"
},
"Prioritization, Recommendations & Actionables": {
"High-Risk Alert Priority": "Low",
"Recommended Actions": {
"Surveillance": "No further surveillance needed.",
"Intervention": "Not required.",
"Humanitarian Assistance": "Not required.",
"Follow-Up Analysis": "No unusual phrases detected requiring review."
}
}
}
```
# Notes
- Ensure that you mark any ambiguous segments as requiring further investigation.
- Pay attention to emotional tone shifts or sudden changes in behavior.
- If any direct or implied threat is detected, prioritize appropriately using the provided classifications.
- Err on the side of caution. In case there is even a remote possibility that there might be something that required human attention, flag it.
"""},
{"role": "user", "content": f"Context: {text}\n\nQuestion: {question}"}
]
response = qa_pipeline(messages, max_new_tokens=256)[0]['generated_text']
return response
except Exception as e:
logger.error(f"Q&A failed: {str(e)}")
return f"Error occurred during Q&A process: {str(e)}"
# Create Gradio interface
with gr.Blocks() as iface:
gr.Markdown("# Automatic Speech Recognition for Indic Languages")
with gr.Row():
with gr.Column():
audio_input = gr.Audio(type="filepath")
translate_checkbox = gr.Checkbox(label="Enable Translation")
process_button = gr.Button("Process Audio")
with gr.Column():
# ASR_RESULT = gr.Textbox(label="Output")
full_text_output = gr.Textbox(label="Full Text", lines=5)
translation_output = gr.Textbox(label="Transcription/Translation", lines=10)
with gr.Row():
with gr.Column():
summarize_button = gr.Button("Summarize")
summary_output = gr.Textbox(label="Summary", lines=3)
with gr.Column():
question_input = gr.Textbox(label="Ask a question about the transcription")
answer_button = gr.Button("Get Answer")
answer_output = gr.Textbox(label="Answer", lines=3)
# Set up event handlers
process_button.click(
process_audio,
inputs=[audio_input, translate_checkbox],
outputs=[translation_output, full_text_output]
# outputs=[ASR_RESULT]
)
# translated_text = ''.join(item['translated'] for item in ASR_RESULT if 'translated' in item)
summarize_button.click(
summarize_text,
# inputs=[ASR_RESULT],
inputs=[translation_output],
outputs=[summary_output]
)
answer_button.click(
answer_question,
inputs=[full_text_output, question_input],
outputs=[answer_output]
)
# Add system information
gr.Markdown(f"""
## System Information
- Device: {"CUDA" if torch.cuda.is_available() else "CPU"}
- CUDA Available: {"Yes" if torch.cuda.is_available() else "No"}
## Features
- Automatic language detection
- High-quality transcription using MMS
- Optional translation to English
- Text summarization
- Question answering
""")
if __name__ == "__main__":
iface.launch(server_port=None) |