Spaces:
Runtime error
Runtime error
Upload 4 files
Browse files- services/audio_error_handling.py +52 -0
- services/llm_service.py +108 -89
- services/resume_service.py +146 -144
- services/voice_service.py +278 -0
services/audio_error_handling.py
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import logging
|
| 2 |
+
import sys
|
| 3 |
+
from pathlib import Path
|
| 4 |
+
|
| 5 |
+
# Configure logging
|
| 6 |
+
LOG_FORMAT = '%(asctime)s - %(name)s - %(levelname)s - %(message)s'
|
| 7 |
+
LOG_LEVEL = logging.INFO
|
| 8 |
+
LOG_DIR = Path(__file__).parent.parent / 'logs'
|
| 9 |
+
LOG_DIR.mkdir(exist_ok=True)
|
| 10 |
+
|
| 11 |
+
# File handler
|
| 12 |
+
file_handler = logging.FileHandler(LOG_DIR / 'audio_processing.log')
|
| 13 |
+
file_handler.setFormatter(logging.Formatter(LOG_FORMAT))
|
| 14 |
+
|
| 15 |
+
# Console handler
|
| 16 |
+
console_handler = logging.StreamHandler(sys.stdout)
|
| 17 |
+
console_handler.setFormatter(logging.Formatter(LOG_FORMAT))
|
| 18 |
+
|
| 19 |
+
# Create logger
|
| 20 |
+
logger = logging.getLogger('audio_service')
|
| 21 |
+
logger.setLevel(LOG_LEVEL)
|
| 22 |
+
logger.addHandler(file_handler)
|
| 23 |
+
logger.addHandler(console_handler)
|
| 24 |
+
|
| 25 |
+
class AudioProcessingError(Exception):
|
| 26 |
+
"""Base exception for audio processing errors"""
|
| 27 |
+
pass
|
| 28 |
+
|
| 29 |
+
class AudioTranscriptionError(AudioProcessingError):
|
| 30 |
+
"""Exception raised for errors during audio transcription"""
|
| 31 |
+
pass
|
| 32 |
+
|
| 33 |
+
class AudioAnalysisError(AudioProcessingError):
|
| 34 |
+
"""Exception raised for errors during audio analysis"""
|
| 35 |
+
pass
|
| 36 |
+
|
| 37 |
+
class AudioFormatError(AudioProcessingError):
|
| 38 |
+
"""Exception raised for unsupported audio formats"""
|
| 39 |
+
pass
|
| 40 |
+
|
| 41 |
+
def handle_audio_error(func):
|
| 42 |
+
"""Decorator for handling audio processing errors"""
|
| 43 |
+
def wrapper(*args, **kwargs):
|
| 44 |
+
try:
|
| 45 |
+
return func(*args, **kwargs)
|
| 46 |
+
except AudioProcessingError as e:
|
| 47 |
+
logger.error(f"Audio processing error in {func.__name__}: {str(e)}")
|
| 48 |
+
raise
|
| 49 |
+
except Exception as e:
|
| 50 |
+
logger.error(f"Unexpected error in {func.__name__}: {str(e)}", exc_info=True)
|
| 51 |
+
raise AudioProcessingError(f"An unexpected error occurred: {str(e)}")
|
| 52 |
+
return wrapper
|
services/llm_service.py
CHANGED
|
@@ -1,89 +1,108 @@
|
|
| 1 |
-
from langchain_google_genai import ChatGoogleGenerativeAI
|
| 2 |
-
from langchain.prompts import PromptTemplate
|
| 3 |
-
from config.settings import get_settings
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
def get_llm():
|
| 7 |
-
settings = get_settings()
|
| 8 |
-
return ChatGoogleGenerativeAI(
|
| 9 |
-
model=settings["llm_model"],
|
| 10 |
-
temperature=settings["temperature"],
|
| 11 |
-
google_api_key=settings["google_api_key"],
|
| 12 |
-
)
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
def get_unified_resume_prompt():
|
| 16 |
-
return PromptTemplate(
|
| 17 |
-
input_variables=["resume_text", "relevant_content", "job_role"],
|
| 18 |
-
template='''You are a professional resume analyzer with expertise in technical recruiting.
|
| 19 |
-
|
| 20 |
-
Your task is to extract both basic information and relevant work experience from the resume.
|
| 21 |
-
|
| 22 |
-
RESUME CONTENT:
|
| 23 |
-
{resume_text}
|
| 24 |
-
|
| 25 |
-
MOST RELEVANT SECTIONS (semantically similar to job requirements):
|
| 26 |
-
{relevant_content}
|
| 27 |
-
|
| 28 |
-
TARGET JOB ROLE/DESCRIPTION: {job_role}
|
| 29 |
-
|
| 30 |
-
TASK INSTRUCTIONS:
|
| 31 |
-
Basic Information:
|
| 32 |
-
1. Extract all personally identifiable information that is explicitly stated in the text
|
| 33 |
-
2. For fields where information is not found, use null
|
| 34 |
-
3. Do not invent or infer information not present in the text
|
| 35 |
-
4. For education, only return the highest level completed (postgraduate > undergraduate > 12th > 10th)
|
| 36 |
-
5. Extract the university/institution name for the highest education qualification
|
| 37 |
-
6. For role, use the most recent/current position mentioned
|
| 38 |
-
7. Return complete values - do not truncate or summarize unnecessarily
|
| 39 |
-
8. Maintain exact formatting/capitalization as shown in the resume
|
| 40 |
-
9. Extract all contact information including email and phone number (just the primary number)
|
| 41 |
-
10. Look for any government ID numbers (PAN, Aadhar, Passport) that might be present
|
| 42 |
-
|
| 43 |
-
Experience Analysis:
|
| 44 |
-
1. Create a single concise paragraph summarizing ALL relevant professional experience that directly relates to the job role
|
| 45 |
-
2. Focus only on experiences that explicitly demonstrate use of the exact technologies/tools required
|
| 46 |
-
3. If NO relevant experience is found, return null for relevant_experience
|
| 47 |
-
4. Provide a brief summary analyzing how well the candidate's experience matches the job role
|
| 48 |
-
|
| 49 |
-
Skill Scoring:
|
| 50 |
-
1. Identify all technical and professional skills mentioned in the resume
|
| 51 |
-
2. For each skill, count its occurrences:
|
| 52 |
-
- If mentioned under projects or experience sections: 5 points per occurrence
|
| 53 |
-
- If mentioned outside projects/experience sections: 3 points per occurrence
|
| 54 |
-
3. Calculate total score for each skill by summing up all occurrences
|
| 55 |
-
4. Include both the skill name and its calculated score
|
| 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 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from langchain_google_genai import ChatGoogleGenerativeAI
|
| 2 |
+
from langchain.prompts import PromptTemplate
|
| 3 |
+
from config.settings import get_settings
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
def get_llm():
|
| 7 |
+
settings = get_settings()
|
| 8 |
+
return ChatGoogleGenerativeAI(
|
| 9 |
+
model=settings["llm_model"],
|
| 10 |
+
temperature=settings["temperature"],
|
| 11 |
+
google_api_key=settings["google_api_key"],
|
| 12 |
+
)
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def get_unified_resume_prompt():
|
| 16 |
+
return PromptTemplate(
|
| 17 |
+
input_variables=["resume_text", "relevant_content", "job_role"],
|
| 18 |
+
template='''You are a professional resume analyzer with expertise in technical recruiting.
|
| 19 |
+
|
| 20 |
+
Your task is to extract both basic information and relevant work experience from the resume.
|
| 21 |
+
|
| 22 |
+
RESUME CONTENT:
|
| 23 |
+
{resume_text}
|
| 24 |
+
|
| 25 |
+
MOST RELEVANT SECTIONS (semantically similar to job requirements):
|
| 26 |
+
{relevant_content}
|
| 27 |
+
|
| 28 |
+
TARGET JOB ROLE/DESCRIPTION: {job_role}
|
| 29 |
+
|
| 30 |
+
TASK INSTRUCTIONS:
|
| 31 |
+
Basic Information:
|
| 32 |
+
1. Extract all personally identifiable information that is explicitly stated in the text
|
| 33 |
+
2. For fields where information is not found, use null
|
| 34 |
+
3. Do not invent or infer information not present in the text
|
| 35 |
+
4. For education, only return the highest level completed (postgraduate > undergraduate > 12th > 10th)
|
| 36 |
+
5. Extract the university/institution name for the highest education qualification
|
| 37 |
+
6. For role, use the most recent/current position mentioned
|
| 38 |
+
7. Return complete values - do not truncate or summarize unnecessarily
|
| 39 |
+
8. Maintain exact formatting/capitalization as shown in the resume
|
| 40 |
+
9. Extract all contact information including email and phone number (just the primary number)
|
| 41 |
+
10. Look for any government ID numbers (PAN, Aadhar, Passport) that might be present
|
| 42 |
+
|
| 43 |
+
Experience Analysis:
|
| 44 |
+
1. Create a single concise paragraph summarizing ALL relevant professional experience that directly relates to the job role
|
| 45 |
+
2. Focus only on experiences that explicitly demonstrate use of the exact technologies/tools required
|
| 46 |
+
3. If NO relevant experience is found, return null for relevant_experience
|
| 47 |
+
4. Provide a brief summary analyzing how well the candidate's experience matches the job role
|
| 48 |
+
|
| 49 |
+
Skill Scoring:
|
| 50 |
+
1. Identify all technical and professional skills mentioned in the resume which are relevant to the job role provided
|
| 51 |
+
2. For each skill, count its occurrences:
|
| 52 |
+
- If mentioned under projects or experience sections: 5 points per occurrence
|
| 53 |
+
- If mentioned outside projects/experience sections: 3 points per occurrence
|
| 54 |
+
3. Calculate total score for each skill by summing up all occurrences
|
| 55 |
+
4. Include both the skill name and its calculated score
|
| 56 |
+
|
| 57 |
+
HR Questions Generation:
|
| 58 |
+
1. Generate as many relevant interview questions based on the candidate's experience and the job role
|
| 59 |
+
2. For each question, provide the expected answer based on information found in the resume
|
| 60 |
+
3. Focus questions on:
|
| 61 |
+
- Technical skills and their practical application
|
| 62 |
+
- Project experiences and challenges
|
| 63 |
+
- Role-specific competencies
|
| 64 |
+
- Career transitions or gaps if any
|
| 65 |
+
4. Questions should be specific to this candidate's background
|
| 66 |
+
5. Answers should strictly reflect information available in the resume
|
| 67 |
+
|
| 68 |
+
FORMAT YOUR RESPONSE AS A VALID JSON OBJECT WITH THIS EXACT STRUCTURE:
|
| 69 |
+
{{
|
| 70 |
+
"candidate_name": "full name of the candidate",
|
| 71 |
+
"gender": "gender if mentioned",
|
| 72 |
+
"date_of_birth": "DOB in YYYY/MM/DD format if mentioned",
|
| 73 |
+
"email": "email address if mentioned",
|
| 74 |
+
"mobile_number": "primary phone number if mentioned",
|
| 75 |
+
"address": "complete physical address if mentioned",
|
| 76 |
+
"pan_number": "PAN number if mentioned",
|
| 77 |
+
"aadhar_number": "Aadhar number if mentioned",
|
| 78 |
+
"passport_number": "Passport number if mentioned",
|
| 79 |
+
"highest_education": "most advanced completed education level, interpret from the data and return only 4 of these options : 10th, 12th, Undergraduate, PostGraduate",
|
| 80 |
+
"university_name": "name of the university/institution where highest education was completed",
|
| 81 |
+
"designation": "current or most recent role/position",
|
| 82 |
+
"skills": "comma-separated list of technical and professional skills",
|
| 83 |
+
"skill_score": [
|
| 84 |
+
{{"skill_name": "name of the skill", "skill_score": numerical_score}},
|
| 85 |
+
...
|
| 86 |
+
],
|
| 87 |
+
"total_experience": "total work experience in terms of years i.e., if experience is 6 months return "0.5", divide the number of months found by 12 and return the number as a string",
|
| 88 |
+
"region": "geographical region/city of work",
|
| 89 |
+
"relevant_experience": "single paragraph summarizing all experience relevant to the job role. If no relevant experience found,reply with a sentence describing they have no relevant experience",
|
| 90 |
+
"experience_summary": "brief analysis of how well the experience matches the job role",
|
| 91 |
+
"hr_questions": [
|
| 92 |
+
{{
|
| 93 |
+
"question": "relevant interview question based on resume content and job role",
|
| 94 |
+
"answer": "expected answer based on the resume content"
|
| 95 |
+
}},
|
| 96 |
+
...
|
| 97 |
+
]
|
| 98 |
+
}}
|
| 99 |
+
|
| 100 |
+
IMPORTANT:
|
| 101 |
+
- DO NOT INVENT OR ADD INFORMATION that is not in the resume
|
| 102 |
+
- BE EXTREMELY STRICT about experience relevance
|
| 103 |
+
- Response must be a valid JSON object
|
| 104 |
+
- For sensitive information like Aadhar/PAN, return only if explicitly mentioned in resume
|
| 105 |
+
- Skills should be returned as a comma-separated string (e.g. "Java, Spring Boot, Docker, AWS, PostgreSQL")
|
| 106 |
+
- Calculate skill scores accurately based on occurrences in projects/experience (5 points) vs elsewhere (3 points)
|
| 107 |
+
- HR questions should be relevant to both the candidate's experience and the target job role'''
|
| 108 |
+
)
|
services/resume_service.py
CHANGED
|
@@ -1,145 +1,147 @@
|
|
| 1 |
-
import json
|
| 2 |
-
from typing import Dict
|
| 3 |
-
from datetime import datetime
|
| 4 |
-
|
| 5 |
-
from core.doc_processor import (
|
| 6 |
-
extract_text_from_bytes,
|
| 7 |
-
extract_text_from_docx,
|
| 8 |
-
parse_pdf_date
|
| 9 |
-
)
|
| 10 |
-
from core.text_processor import chunk_text
|
| 11 |
-
from core.vector_store import create_or_load_vectorstore, similarity_search
|
| 12 |
-
from services.llm_service import get_llm, get_unified_resume_prompt
|
| 13 |
-
from services.db_service import store_candidate_data
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
async def analyze_resume_document(
|
| 17 |
-
file_bytes: bytes, job_role: str, file_extension: str
|
| 18 |
-
) -> Dict:
|
| 19 |
-
if file_extension.lower() == ".pdf":
|
| 20 |
-
resume_text, mod_date = extract_text_from_bytes(file_bytes)
|
| 21 |
-
file_modified_date = parse_pdf_date(mod_date) if mod_date else datetime.now()
|
| 22 |
-
elif file_extension.lower() == ".docx":
|
| 23 |
-
resume_text, mod_date = extract_text_from_docx(file_bytes)
|
| 24 |
-
file_modified_date = mod_date if mod_date else datetime.now()
|
| 25 |
-
else:
|
| 26 |
-
raise ValueError(f"Unsupported file format: {file_extension}")
|
| 27 |
-
|
| 28 |
-
if resume_text.startswith("Error"):
|
| 29 |
-
raise ValueError(resume_text)
|
| 30 |
-
|
| 31 |
-
chunks = chunk_text(resume_text)
|
| 32 |
-
vector_store = create_or_load_vectorstore(chunks)
|
| 33 |
-
|
| 34 |
-
analysis_result = analyze_resume_with_llm(vector_store, resume_text, job_role, file_modified_date)
|
| 35 |
-
|
| 36 |
-
try:
|
| 37 |
-
db_record = await store_candidate_data(analysis_result)
|
| 38 |
-
analysis_result["id"] = db_record.get("id")
|
| 39 |
-
analysis_result["stored_in_db"] = True
|
| 40 |
-
except Exception as e:
|
| 41 |
-
print(f"Error storing candidate data: {str(e)}")
|
| 42 |
-
analysis_result["stored_in_db"] = False
|
| 43 |
-
analysis_result["db_error"] = str(e)
|
| 44 |
-
|
| 45 |
-
return analysis_result
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
def analyze_resume_with_llm(vectorstore, resume_text: str, job_role: str, file_modified_date: datetime) -> Dict:
|
| 49 |
-
try:
|
| 50 |
-
retrieved_docs = similarity_search(vectorstore, job_role)
|
| 51 |
-
relevant_content = "\n".join([doc.page_content for doc in retrieved_docs])
|
| 52 |
-
|
| 53 |
-
unified_prompt = get_unified_resume_prompt()
|
| 54 |
-
llm = get_llm()
|
| 55 |
-
chain = unified_prompt | llm
|
| 56 |
-
|
| 57 |
-
response = chain.invoke(
|
| 58 |
-
{
|
| 59 |
-
"resume_text": resume_text,
|
| 60 |
-
"relevant_content": relevant_content,
|
| 61 |
-
"job_role": job_role,
|
| 62 |
-
}
|
| 63 |
-
)
|
| 64 |
-
|
| 65 |
-
response_text = response.content.strip()
|
| 66 |
-
if response_text.startswith("`"):
|
| 67 |
-
response_text = response_text.strip("`")
|
| 68 |
-
if response_text.startswith("json"):
|
| 69 |
-
response_text = response_text[4:]
|
| 70 |
-
|
| 71 |
-
parsed_response = json.loads(response_text.strip())
|
| 72 |
-
|
| 73 |
-
default_response = {
|
| 74 |
-
"candidate_name": "",
|
| 75 |
-
"gender": None,
|
| 76 |
-
"date_of_birth": None,
|
| 77 |
-
"email": None,
|
| 78 |
-
"mobile_number": None,
|
| 79 |
-
"address": None,
|
| 80 |
-
"pan_number": None,
|
| 81 |
-
"aadhar_number": None,
|
| 82 |
-
"passport_number": None,
|
| 83 |
-
"highest_education": None,
|
| 84 |
-
"university_name": None,
|
| 85 |
-
"designation": None,
|
| 86 |
-
"skills": "",
|
| 87 |
-
"skill_score": [],
|
| 88 |
-
"total_experience": None,
|
| 89 |
-
"region": None,
|
| 90 |
-
"relevant_experience": None,
|
| 91 |
-
"experience_summary": "No relevant experience analysis available",
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
result["file_modified_date"] = file_modified_date
|
| 97 |
-
|
| 98 |
-
return result
|
| 99 |
-
|
| 100 |
-
except json.JSONDecodeError as e:
|
| 101 |
-
print(f"JSON Parse Error: {str(e)}")
|
| 102 |
-
return {
|
| 103 |
-
"candidate_name": "",
|
| 104 |
-
"gender": None,
|
| 105 |
-
"date_of_birth": None,
|
| 106 |
-
"email": None,
|
| 107 |
-
"mobile_number": None,
|
| 108 |
-
"address": None,
|
| 109 |
-
"pan_number": None,
|
| 110 |
-
"aadhar_number": None,
|
| 111 |
-
"passport_number": None,
|
| 112 |
-
"highest_education": None,
|
| 113 |
-
"university_name": None,
|
| 114 |
-
"designation": None,
|
| 115 |
-
"skills": "",
|
| 116 |
-
"skill_score": [],
|
| 117 |
-
"total_experience": None,
|
| 118 |
-
"region": None,
|
| 119 |
-
"relevant_experience": None,
|
| 120 |
-
"experience_summary": "Could not parse resume details from response",
|
| 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 |
}
|
|
|
|
| 1 |
+
import json
|
| 2 |
+
from typing import Dict
|
| 3 |
+
from datetime import datetime
|
| 4 |
+
|
| 5 |
+
from core.doc_processor import (
|
| 6 |
+
extract_text_from_bytes,
|
| 7 |
+
extract_text_from_docx,
|
| 8 |
+
parse_pdf_date
|
| 9 |
+
)
|
| 10 |
+
from core.text_processor import chunk_text
|
| 11 |
+
from core.vector_store import create_or_load_vectorstore, similarity_search
|
| 12 |
+
from services.llm_service import get_llm, get_unified_resume_prompt
|
| 13 |
+
from services.db_service import store_candidate_data
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
async def analyze_resume_document(
|
| 17 |
+
file_bytes: bytes, job_role: str, file_extension: str
|
| 18 |
+
) -> Dict:
|
| 19 |
+
if file_extension.lower() == ".pdf":
|
| 20 |
+
resume_text, mod_date = extract_text_from_bytes(file_bytes)
|
| 21 |
+
file_modified_date = parse_pdf_date(mod_date) if mod_date else datetime.now()
|
| 22 |
+
elif file_extension.lower() == ".docx":
|
| 23 |
+
resume_text, mod_date = extract_text_from_docx(file_bytes)
|
| 24 |
+
file_modified_date = mod_date if mod_date else datetime.now()
|
| 25 |
+
else:
|
| 26 |
+
raise ValueError(f"Unsupported file format: {file_extension}")
|
| 27 |
+
|
| 28 |
+
if resume_text.startswith("Error"):
|
| 29 |
+
raise ValueError(resume_text)
|
| 30 |
+
|
| 31 |
+
chunks = chunk_text(resume_text)
|
| 32 |
+
vector_store = create_or_load_vectorstore(chunks)
|
| 33 |
+
|
| 34 |
+
analysis_result = analyze_resume_with_llm(vector_store, resume_text, job_role, file_modified_date)
|
| 35 |
+
|
| 36 |
+
try:
|
| 37 |
+
db_record = await store_candidate_data(analysis_result)
|
| 38 |
+
analysis_result["id"] = db_record.get("id")
|
| 39 |
+
analysis_result["stored_in_db"] = True
|
| 40 |
+
except Exception as e:
|
| 41 |
+
print(f"Error storing candidate data: {str(e)}")
|
| 42 |
+
analysis_result["stored_in_db"] = False
|
| 43 |
+
analysis_result["db_error"] = str(e)
|
| 44 |
+
|
| 45 |
+
return analysis_result
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
def analyze_resume_with_llm(vectorstore, resume_text: str, job_role: str, file_modified_date: datetime) -> Dict:
|
| 49 |
+
try:
|
| 50 |
+
retrieved_docs = similarity_search(vectorstore, job_role)
|
| 51 |
+
relevant_content = "\n".join([doc.page_content for doc in retrieved_docs])
|
| 52 |
+
|
| 53 |
+
unified_prompt = get_unified_resume_prompt()
|
| 54 |
+
llm = get_llm()
|
| 55 |
+
chain = unified_prompt | llm
|
| 56 |
+
|
| 57 |
+
response = chain.invoke(
|
| 58 |
+
{
|
| 59 |
+
"resume_text": resume_text,
|
| 60 |
+
"relevant_content": relevant_content,
|
| 61 |
+
"job_role": job_role,
|
| 62 |
+
}
|
| 63 |
+
)
|
| 64 |
+
|
| 65 |
+
response_text = response.content.strip()
|
| 66 |
+
if response_text.startswith("`"):
|
| 67 |
+
response_text = response_text.strip("`")
|
| 68 |
+
if response_text.startswith("json"):
|
| 69 |
+
response_text = response_text[4:]
|
| 70 |
+
|
| 71 |
+
parsed_response = json.loads(response_text.strip())
|
| 72 |
+
|
| 73 |
+
default_response = {
|
| 74 |
+
"candidate_name": "",
|
| 75 |
+
"gender": None,
|
| 76 |
+
"date_of_birth": None,
|
| 77 |
+
"email": None,
|
| 78 |
+
"mobile_number": None,
|
| 79 |
+
"address": None,
|
| 80 |
+
"pan_number": None,
|
| 81 |
+
"aadhar_number": None,
|
| 82 |
+
"passport_number": None,
|
| 83 |
+
"highest_education": None,
|
| 84 |
+
"university_name": None,
|
| 85 |
+
"designation": None,
|
| 86 |
+
"skills": "",
|
| 87 |
+
"skill_score": [],
|
| 88 |
+
"total_experience": None,
|
| 89 |
+
"region": None,
|
| 90 |
+
"relevant_experience": None,
|
| 91 |
+
"experience_summary": "No relevant experience analysis available",
|
| 92 |
+
"hr_questions": []
|
| 93 |
+
}
|
| 94 |
+
|
| 95 |
+
result = {**default_response, **parsed_response}
|
| 96 |
+
result["file_modified_date"] = file_modified_date
|
| 97 |
+
|
| 98 |
+
return result
|
| 99 |
+
|
| 100 |
+
except json.JSONDecodeError as e:
|
| 101 |
+
print(f"JSON Parse Error: {str(e)}")
|
| 102 |
+
return {
|
| 103 |
+
"candidate_name": "",
|
| 104 |
+
"gender": None,
|
| 105 |
+
"date_of_birth": None,
|
| 106 |
+
"email": None,
|
| 107 |
+
"mobile_number": None,
|
| 108 |
+
"address": None,
|
| 109 |
+
"pan_number": None,
|
| 110 |
+
"aadhar_number": None,
|
| 111 |
+
"passport_number": None,
|
| 112 |
+
"highest_education": None,
|
| 113 |
+
"university_name": None,
|
| 114 |
+
"designation": None,
|
| 115 |
+
"skills": "",
|
| 116 |
+
"skill_score": [],
|
| 117 |
+
"total_experience": None,
|
| 118 |
+
"region": None,
|
| 119 |
+
"relevant_experience": None,
|
| 120 |
+
"experience_summary": "Could not parse resume details from response",
|
| 121 |
+
"hr_questions": [],
|
| 122 |
+
"file_modified_date": file_modified_date
|
| 123 |
+
}
|
| 124 |
+
except Exception as e:
|
| 125 |
+
print(f"Unexpected Error: {str(e)}")
|
| 126 |
+
return {
|
| 127 |
+
"candidate_name": "",
|
| 128 |
+
"gender": None,
|
| 129 |
+
"date_of_birth": None,
|
| 130 |
+
"email": None,
|
| 131 |
+
"mobile_number": None,
|
| 132 |
+
"address": None,
|
| 133 |
+
"pan_number": None,
|
| 134 |
+
"aadhar_number": None,
|
| 135 |
+
"passport_number": None,
|
| 136 |
+
"highest_education": None,
|
| 137 |
+
"university_name": None,
|
| 138 |
+
"designation": None,
|
| 139 |
+
"skills": "",
|
| 140 |
+
"skill_score": [],
|
| 141 |
+
"total_experience": None,
|
| 142 |
+
"region": None,
|
| 143 |
+
"relevant_experience": None,
|
| 144 |
+
"experience_summary": f"Error analyzing resume: {str(e)}",
|
| 145 |
+
"hr_questions": [],
|
| 146 |
+
"file_modified_date": file_modified_date
|
| 147 |
}
|
services/voice_service.py
ADDED
|
@@ -0,0 +1,278 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import io
|
| 2 |
+
import os
|
| 3 |
+
from pathlib import Path
|
| 4 |
+
from typing import Dict, Tuple
|
| 5 |
+
import librosa
|
| 6 |
+
import numpy as np
|
| 7 |
+
import speech_recognition as sr
|
| 8 |
+
import soundfile as sf
|
| 9 |
+
from .llm_service import get_llm
|
| 10 |
+
from langchain.prompts import PromptTemplate
|
| 11 |
+
|
| 12 |
+
from .audio_error_handling import (
|
| 13 |
+
handle_audio_error, AudioProcessingError, AudioFormatError,
|
| 14 |
+
AudioAnalysisError, logger
|
| 15 |
+
)
|
| 16 |
+
|
| 17 |
+
# Constants
|
| 18 |
+
MAX_FILE_SIZE = 25 * 1024 * 1024 # 25MB
|
| 19 |
+
SUPPORTED_FORMATS = {'.wav', '.mp3', '.m4a'}
|
| 20 |
+
MAX_DURATION = 300 # 5 minutes in seconds
|
| 21 |
+
|
| 22 |
+
class AudioAnalyzer:
|
| 23 |
+
def __init__(self):
|
| 24 |
+
self.recognizer = sr.Recognizer()
|
| 25 |
+
|
| 26 |
+
@handle_audio_error
|
| 27 |
+
def validate_audio_file(self, file_path: Path) -> None:
|
| 28 |
+
"""Validate audio file format and size"""
|
| 29 |
+
if not file_path.suffix.lower() in SUPPORTED_FORMATS:
|
| 30 |
+
raise AudioFormatError(f"Unsupported audio format. Supported formats: {', '.join(SUPPORTED_FORMATS)}")
|
| 31 |
+
|
| 32 |
+
if os.path.getsize(file_path) > MAX_FILE_SIZE:
|
| 33 |
+
raise AudioFormatError(f"File size exceeds maximum limit of {MAX_FILE_SIZE/1024/1024:.1f}MB")
|
| 34 |
+
|
| 35 |
+
@handle_audio_error
|
| 36 |
+
def _extract_audio_features(self, audio_data: bytes) -> Tuple[np.ndarray, float]:
|
| 37 |
+
"""Extract audio features using librosa with error handling"""
|
| 38 |
+
try:
|
| 39 |
+
audio_array, sample_rate = sf.read(io.BytesIO(audio_data))
|
| 40 |
+
|
| 41 |
+
# If stereo, convert to mono
|
| 42 |
+
if len(audio_array.shape) > 1:
|
| 43 |
+
audio_array = np.mean(audio_array, axis=1)
|
| 44 |
+
|
| 45 |
+
# Check duration
|
| 46 |
+
duration = len(audio_array) / sample_rate
|
| 47 |
+
if duration > MAX_DURATION:
|
| 48 |
+
raise AudioFormatError(f"Audio duration exceeds maximum limit of {MAX_DURATION/60:.1f} minutes")
|
| 49 |
+
|
| 50 |
+
return audio_array, sample_rate
|
| 51 |
+
except Exception as e:
|
| 52 |
+
logger.error(f"Error extracting audio features: {e}")
|
| 53 |
+
raise AudioProcessingError("Failed to process audio file")
|
| 54 |
+
|
| 55 |
+
def _transcribe_audio(self, audio_array: np.ndarray, sample_rate: float) -> str:
|
| 56 |
+
"""Transcribe audio using speech recognition"""
|
| 57 |
+
# Convert numpy array to audio file format recognized by speech_recognition
|
| 58 |
+
with io.BytesIO() as wav_io:
|
| 59 |
+
sf.write(wav_io, audio_array, sample_rate, format='wav')
|
| 60 |
+
wav_io.seek(0)
|
| 61 |
+
with sr.AudioFile(wav_io) as source:
|
| 62 |
+
audio_data = self.recognizer.record(source)
|
| 63 |
+
try:
|
| 64 |
+
return self.recognizer.recognize_google(audio_data)
|
| 65 |
+
except sr.UnknownValueError:
|
| 66 |
+
return "Speech Recognition could not understand the audio"
|
| 67 |
+
except sr.RequestError:
|
| 68 |
+
return "Could not request results from Speech Recognition service"
|
| 69 |
+
|
| 70 |
+
def _calculate_fluency_score(self, audio_array: np.ndarray, sample_rate: float) -> float:
|
| 71 |
+
"""Calculate fluency score based on silence intervals and speech rate"""
|
| 72 |
+
# Get onsets for words
|
| 73 |
+
onset_env = librosa.onset.onset_strength(y=audio_array, sr=sample_rate)
|
| 74 |
+
tempo, _ = librosa.beat.beat_track(onset_envelope=onset_env, sr=sample_rate)
|
| 75 |
+
|
| 76 |
+
# Calculate speech rate normalized to a 0-1 scale
|
| 77 |
+
optimal_tempo = 150 # Increased from 120 to be more lenient
|
| 78 |
+
tempo_score = 1 - min(1.0, abs(tempo - optimal_tempo) / optimal_tempo)
|
| 79 |
+
|
| 80 |
+
# Calculate silence ratio with more lenient threshold
|
| 81 |
+
silence_threshold = 0.02 # Reduced from 0.03 to be more lenient
|
| 82 |
+
is_silence = audio_array < silence_threshold
|
| 83 |
+
silence_ratio = np.mean(is_silence)
|
| 84 |
+
|
| 85 |
+
# Combine metrics with adjusted weights
|
| 86 |
+
fluency_score = (tempo_score * 0.8 + (1 - silence_ratio) * 0.2) # Increased weight on tempo
|
| 87 |
+
return min(1.0, max(0.0, fluency_score))
|
| 88 |
+
|
| 89 |
+
def _analyze_tone(self, transcribed_text: str) -> Dict:
|
| 90 |
+
"""Analyze tone based on text characteristics"""
|
| 91 |
+
# Analyze based on punctuation and sentence structure
|
| 92 |
+
sentences = [s.strip() for s in transcribed_text.split('.') if s.strip()]
|
| 93 |
+
avg_sentence_length = np.mean([len(s.split()) for s in sentences]) if sentences else 0
|
| 94 |
+
|
| 95 |
+
# Simple tone estimation based on sentence characteristics
|
| 96 |
+
tone_score = min(1.0, avg_sentence_length / 20) # Normalize to 0-1
|
| 97 |
+
return {
|
| 98 |
+
"overall_tone": float(tone_score)
|
| 99 |
+
}
|
| 100 |
+
|
| 101 |
+
def _calculate_confidence_score(self, audio_array: np.ndarray, transcribed_text: str) -> float:
|
| 102 |
+
"""Calculate confidence score based on audio characteristics and text"""
|
| 103 |
+
# Audio energy variation with more lenient thresholds
|
| 104 |
+
rmse = librosa.feature.rms(y=audio_array)[0]
|
| 105 |
+
energy_std = np.std(rmse)
|
| 106 |
+
energy_score = min(1.0, energy_std * 20)
|
| 107 |
+
|
| 108 |
+
# Speech rate consistency with adjusted parameters
|
| 109 |
+
onset_env = librosa.onset.onset_strength(y=audio_array)
|
| 110 |
+
onset_std = np.std(onset_env)
|
| 111 |
+
consistency_score = 1 - min(1.0, onset_std / 10) # was 8
|
| 112 |
+
|
| 113 |
+
|
| 114 |
+
# Combine scores with adjusted weights
|
| 115 |
+
confidence_score = (energy_score * 0.7 + consistency_score * 0.3)
|
| 116 |
+
return min(1.0, max(0.0, confidence_score))
|
| 117 |
+
|
| 118 |
+
def _get_grammar_score(self, text: str) -> float:
|
| 119 |
+
"""
|
| 120 |
+
Calculate a basic grammar score based on sentence structure
|
| 121 |
+
Note: This is a simplified version. For production, use a proper grammar checker API
|
| 122 |
+
"""
|
| 123 |
+
sentences = [s.strip() for s in text.split('.') if s.strip()]
|
| 124 |
+
word_counts = [len(s.split()) for s in sentences]
|
| 125 |
+
|
| 126 |
+
# Analyze sentence length variation
|
| 127 |
+
sentence_length_std = np.std(word_counts) if word_counts else 0
|
| 128 |
+
length_score = 1 - min(1.0, sentence_length_std / 15) # higher tolerance for variance
|
| 129 |
+
|
| 130 |
+
# Count basic punctuation
|
| 131 |
+
punctuation_score = min(1.0, (sum(1 for c in text if c in '.,!?') / len(text)) * 2) # amplify good punctuation
|
| 132 |
+
|
| 133 |
+
return min(1.0, (length_score * 0.7 + punctuation_score * 0.3))
|
| 134 |
+
|
| 135 |
+
def _calculate_wpm(self, audio_duration: float, transcribed_text: str) -> float:
|
| 136 |
+
"""Calculate Words Per Minute"""
|
| 137 |
+
word_count = len(transcribed_text.split())
|
| 138 |
+
minutes = audio_duration / 60
|
| 139 |
+
return word_count / minutes if minutes > 0 else 0
|
| 140 |
+
|
| 141 |
+
|
| 142 |
+
def _get_assessment_prompt(self) -> PromptTemplate:
|
| 143 |
+
"""Create a prompt template for LLM assessment"""
|
| 144 |
+
template = '''You are an interview evaluation assistant. Based on the given metrics—fluency, confidence, grammar, tone, and pauses—generate a brief, constructive assessment of the candidate's spoken response. Use a tone that is supportive and human-like, suitable for hiring and recruitment purposes.
|
| 145 |
+
|
| 146 |
+
Audio Analysis Metrics:
|
| 147 |
+
- Fluency Score: {fluency_score}
|
| 148 |
+
- Grammar Score: {grammar_score}
|
| 149 |
+
- Confidence Score: {confidence_score}
|
| 150 |
+
- Clarity Score: {clarity_score}
|
| 151 |
+
- Pronunciation Score: {pronunciation_score}
|
| 152 |
+
- Words Per Minute: {wpm_rate}
|
| 153 |
+
- Filler Words Count: {filler_words_count}
|
| 154 |
+
- Speaking Duration: {speaking_duration}
|
| 155 |
+
- Number of Pauses: {pauses_count}
|
| 156 |
+
|
| 157 |
+
Transcribed Response:
|
| 158 |
+
"{transcribed_text}"
|
| 159 |
+
|
| 160 |
+
Job Role: {job_role}
|
| 161 |
+
|
| 162 |
+
Please provide a comprehensive assessment including:
|
| 163 |
+
1. Overall communication effectiveness
|
| 164 |
+
2. Areas of strength
|
| 165 |
+
3. Areas for improvement
|
| 166 |
+
4. Interview performance score (0-100)
|
| 167 |
+
5. Hiring recommendation (recommend/consider/reject)
|
| 168 |
+
|
| 169 |
+
Always mention:
|
| 170 |
+
1. Strengths: Start by highlighting what the candidate did well.
|
| 171 |
+
2. Areas for improvement: Be gentle, offering actionable suggestions.
|
| 172 |
+
3. Overall impression: Summarize whether the delivery is interview-ready or needs polishing.
|
| 173 |
+
|
| 174 |
+
Be encouraging and avoid sounding robotic or overly critical.
|
| 175 |
+
Format your response as a clear, concise paragraph that covers all these points.
|
| 176 |
+
Focus on being direct and actionable in your assessment.'''
|
| 177 |
+
return PromptTemplate(
|
| 178 |
+
input_variables=[
|
| 179 |
+
"fluency_score", "grammar_score", "confidence_score",
|
| 180 |
+
"clarity_score", "pronunciation_score",
|
| 181 |
+
"wpm_rate", "filler_words_count", "speaking_duration",
|
| 182 |
+
"pauses_count", "transcribed_text", "job_role"
|
| 183 |
+
],
|
| 184 |
+
template=template
|
| 185 |
+
)
|
| 186 |
+
|
| 187 |
+
def _get_llm_assessment(self, metrics: Dict, job_role: str) -> str:
|
| 188 |
+
"""Get LLM assessment of the audio analysis metrics"""
|
| 189 |
+
try:
|
| 190 |
+
llm = get_llm()
|
| 191 |
+
prompt = self._get_assessment_prompt()
|
| 192 |
+
|
| 193 |
+
# Prepare input variables
|
| 194 |
+
input_vars = {
|
| 195 |
+
**{k: metrics[k] for k in [
|
| 196 |
+
"fluency_score", "grammar_score", "confidence_score",
|
| 197 |
+
"clarity_score", "pronunciation_score",
|
| 198 |
+
"wpm_rate", "filler_words_count", "speaking_duration",
|
| 199 |
+
"pauses_count", "transcribed_text"
|
| 200 |
+
]},
|
| 201 |
+
"job_role": job_role or "General"
|
| 202 |
+
}
|
| 203 |
+
|
| 204 |
+
# Get LLM response
|
| 205 |
+
response = llm.invoke(prompt.format_prompt(**input_vars))
|
| 206 |
+
|
| 207 |
+
# Extract just the content from the response
|
| 208 |
+
response_text = response.content.strip()
|
| 209 |
+
if response_text.startswith("`"):
|
| 210 |
+
response_text = response_text.strip("`")
|
| 211 |
+
if response_text.startswith("json"):
|
| 212 |
+
response_text = response_text[4:]
|
| 213 |
+
|
| 214 |
+
return response_text.strip()
|
| 215 |
+
|
| 216 |
+
except Exception as e:
|
| 217 |
+
print(f"Error getting LLM assessment: {str(e)}")
|
| 218 |
+
# Return a basic assessment based on metrics
|
| 219 |
+
return f"""Basic analysis shows {metrics['speaking_duration']:.1f} seconds of speech with fluency score of {metrics['fluency_score']:.2f}.
|
| 220 |
+
Speech clarity is {metrics['clarity_score']:.2f} and technical accuracy is {metrics['technical_accuracy']:.2f}.
|
| 221 |
+
The candidate shows potential but could provide more technical details.
|
| 222 |
+
Interview score: {int(metrics['confidence_score'] * 100)}/100.
|
| 223 |
+
Recommendation: Consider for further evaluation."""
|
| 224 |
+
|
| 225 |
+
def analyze_audio(self, audio_data: bytes, job_role: str = None) -> Dict:
|
| 226 |
+
"""Main method to analyze audio and return a simplified response with transcription and assessment"""
|
| 227 |
+
try:
|
| 228 |
+
# Extract audio features
|
| 229 |
+
audio_array, sample_rate = self._extract_audio_features(audio_data)
|
| 230 |
+
duration = len(audio_array) / sample_rate
|
| 231 |
+
|
| 232 |
+
# Get transcription
|
| 233 |
+
transcribed_text = self._transcribe_audio(audio_array, sample_rate)
|
| 234 |
+
|
| 235 |
+
# Calculate various scores internally
|
| 236 |
+
fluency_score = self._calculate_fluency_score(audio_array, sample_rate)
|
| 237 |
+
grammar_score = self._get_grammar_score(transcribed_text)
|
| 238 |
+
confidence_score = self._calculate_confidence_score(audio_array, transcribed_text)
|
| 239 |
+
tone_analysis = self._analyze_tone(transcribed_text)
|
| 240 |
+
wpm = self._calculate_wpm(duration, transcribed_text)
|
| 241 |
+
|
| 242 |
+
# Calculate clarity score based on audio features with adjusted parameters
|
| 243 |
+
onset_env = librosa.onset.onset_strength(y=audio_array, sr=sample_rate)
|
| 244 |
+
clarity_score = min(1.0, np.mean(onset_env) / 8)
|
| 245 |
+
|
| 246 |
+
# Count pauses with more lenient parameters
|
| 247 |
+
silence_intervals = librosa.effects.split(audio_array, top_db=35)
|
| 248 |
+
pauses_count = len(silence_intervals) - 1
|
| 249 |
+
|
| 250 |
+
# Simplified filler words count with reduced list
|
| 251 |
+
filler_words = ["um", "uh"] # Reduced list to only count significant fillers
|
| 252 |
+
filler_count = sum(transcribed_text.lower().count(word) for word in filler_words)
|
| 253 |
+
|
| 254 |
+
# Create metrics dictionary for LLM assessment
|
| 255 |
+
metrics = {
|
| 256 |
+
"fluency_score": float(fluency_score),
|
| 257 |
+
"grammar_score": float(grammar_score),
|
| 258 |
+
"confidence_score": float(confidence_score),
|
| 259 |
+
"clarity_score": float(clarity_score),
|
| 260 |
+
"pronunciation_score": float(clarity_score * 0.7 + fluency_score * 0.3),
|
| 261 |
+
"transcribed_text": transcribed_text,
|
| 262 |
+
"wpm_rate": float(wpm),
|
| 263 |
+
"filler_words_count": filler_count,
|
| 264 |
+
"speaking_duration": float(duration),
|
| 265 |
+
"pauses_count": pauses_count
|
| 266 |
+
}
|
| 267 |
+
|
| 268 |
+
# Get LLM assessment
|
| 269 |
+
assessment = self._get_llm_assessment(metrics, job_role)
|
| 270 |
+
|
| 271 |
+
# Return simplified response
|
| 272 |
+
return {
|
| 273 |
+
"transcript": transcribed_text,
|
| 274 |
+
"assessment": assessment
|
| 275 |
+
}
|
| 276 |
+
except Exception as e:
|
| 277 |
+
print(f"Error in analyze_audio: {str(e)}")
|
| 278 |
+
raise AudioAnalysisError(f"Failed to analyze audio: {str(e)}")
|