import gradio as gr
import boto3
import json
import re
import logging
import os
import tempfile
import shutil
import time
import uuid
from datetime import datetime
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Try to import ReportLab (needed for PDF generation)
try:
from reportlab.lib.pagesizes import letter
from reportlab.lib import colors
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer, Table, TableStyle
from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle
REPORTLAB_AVAILABLE = True
except ImportError:
logger.warning("ReportLab library not available - PDF export will be disabled")
REPORTLAB_AVAILABLE = False
# AWS credentials for Bedrock API
AWS_ACCESS_KEY = os.getenv("AWS_ACCESS_KEY", "")
AWS_SECRET_KEY = os.getenv("AWS_SECRET_KEY", "")
AWS_REGION = os.getenv("AWS_REGION", "us-east-1")
# Initialize AWS clients if credentials are available
bedrock_client = None
transcribe_client = None
s3_client = None
if AWS_ACCESS_KEY and AWS_SECRET_KEY:
try:
# Initialize Bedrock client for AI analysis
bedrock_client = boto3.client(
'bedrock-runtime',
aws_access_key_id=AWS_ACCESS_KEY,
aws_secret_access_key=AWS_SECRET_KEY,
region_name=AWS_REGION
)
logger.info("Bedrock client initialized successfully")
# Initialize Transcribe client for speech-to-text
transcribe_client = boto3.client(
'transcribe',
aws_access_key_id=AWS_ACCESS_KEY,
aws_secret_access_key=AWS_SECRET_KEY,
region_name=AWS_REGION
)
logger.info("Transcribe client initialized successfully")
# Initialize S3 client for storing audio files
s3_client = boto3.client(
's3',
aws_access_key_id=AWS_ACCESS_KEY,
aws_secret_access_key=AWS_SECRET_KEY,
region_name=AWS_REGION
)
logger.info("S3 client initialized successfully")
except Exception as e:
logger.error(f"Failed to initialize AWS clients: {str(e)}")
# S3 bucket for storing audio files
S3_BUCKET = os.environ.get("S3_BUCKET", "casl-audio-files")
S3_PREFIX = "transcribe-audio/"
# Create data directories if they don't exist
DATA_DIR = os.environ.get("DATA_DIR", "patient_data")
DOWNLOADS_DIR = os.path.join(DATA_DIR, "downloads")
AUDIO_DIR = os.path.join(DATA_DIR, "audio")
def ensure_data_dirs():
"""Ensure data directories exist"""
global DOWNLOADS_DIR, AUDIO_DIR
try:
os.makedirs(DATA_DIR, exist_ok=True)
os.makedirs(DOWNLOADS_DIR, exist_ok=True)
os.makedirs(AUDIO_DIR, exist_ok=True)
logger.info(f"Data directories created: {DATA_DIR}, {DOWNLOADS_DIR}, {AUDIO_DIR}")
except Exception as e:
logger.warning(f"Could not create data directories: {str(e)}")
# Fallback to tmp directory on HF Spaces
DOWNLOADS_DIR = os.path.join(tempfile.gettempdir(), "casl_downloads")
AUDIO_DIR = os.path.join(tempfile.gettempdir(), "casl_audio")
os.makedirs(DOWNLOADS_DIR, exist_ok=True)
os.makedirs(AUDIO_DIR, exist_ok=True)
logger.info(f"Using fallback directories: {DOWNLOADS_DIR}, {AUDIO_DIR}")
# Initialize data directories
ensure_data_dirs()
# Sample transcript for the demo
SAMPLE_TRANSCRIPT = """*PAR: today I would &-um like to talk about &-um a fun trip I took last &-um summer with my family.
*PAR: we went to the &-um &-um beach [//] no to the mountains [//] I mean the beach actually.
*PAR: there was lots of &-um &-um swimming and &-um sun.
*PAR: we [/] we stayed for &-um three no [//] four days in a &-um hotel near the water [: ocean] [*].
*PAR: my favorite part was &-um building &-um castles with sand.
*PAR: sometimes I forget [//] forgetted [: forgot] [*] what they call those things we built.
*PAR: my brother he [//] he helped me dig a big hole.
*PAR: we saw [/] saw fishies [: fish] [*] swimming in the water.
*PAR: sometimes I wonder [/] wonder where fishies [: fish] [*] go when it's cold.
*PAR: maybe they have [/] have houses under the water.
*PAR: after swimming we [//] I eat [: ate] [*] &-um ice cream with &-um chocolate things on top.
*PAR: what do you call those &-um &-um sprinkles! that's the word.
*PAR: my mom said to &-um that I could have &-um two scoops next time.
*PAR: I want to go back to the beach [/] beach next year."""
def read_cha_file(file_path):
"""Read and parse a .cha transcript file"""
try:
with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
content = f.read()
# Extract participant lines (starting with *PAR:)
par_lines = []
for line in content.splitlines():
if line.startswith('*PAR:'):
par_lines.append(line)
# If no PAR lines found, just return the whole content
if not par_lines:
return content
return '\n'.join(par_lines)
except Exception as e:
logger.error(f"Error reading CHA file: {str(e)}")
return ""
def process_upload(file):
"""Process an uploaded file (PDF, text, or CHA)"""
if file is None:
return ""
file_path = file.name
if file_path.endswith('.pdf'):
# For PDF, we would need PyPDF2 or similar
return "PDF upload not supported in this simple version"
elif file_path.endswith('.cha'):
return read_cha_file(file_path)
else:
with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
return f.read()
def call_bedrock(prompt, max_tokens=4096):
"""Call the AWS Bedrock API to analyze text using Claude"""
if not bedrock_client:
return "AWS credentials not configured. Using demo response instead."
try:
body = json.dumps({
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": max_tokens,
"messages": [
{
"role": "user",
"content": prompt
}
],
"temperature": 0.3,
"top_p": 0.9
})
modelId = 'anthropic.claude-3-sonnet-20240229-v1:0'
response = bedrock_client.invoke_model(
body=body,
modelId=modelId,
accept='application/json',
contentType='application/json'
)
response_body = json.loads(response.get('body').read())
return response_body['content'][0]['text']
except Exception as e:
logger.error(f"Error in call_bedrock: {str(e)}")
return f"Error: {str(e)}"
def transcribe_audio(audio_path, patient_age=8):
"""Transcribe an audio recording using Amazon Transcribe and format in CHAT format"""
if not os.path.exists(audio_path):
logger.error(f"Audio file not found: {audio_path}")
return "Error: Audio file not found."
if not transcribe_client or not s3_client:
logger.warning("AWS clients not initialized, using demo transcription")
return generate_demo_transcription()
try:
# Get file info
file_name = os.path.basename(audio_path)
file_size = os.path.getsize(audio_path)
_, file_extension = os.path.splitext(file_name)
# Check file format
supported_formats = ['.mp3', '.mp4', '.wav', '.flac', '.ogg', '.amr', '.webm']
if file_extension.lower() not in supported_formats:
logger.error(f"Unsupported audio format: {file_extension}")
return f"Error: Unsupported audio format. Please use one of: {', '.join(supported_formats)}"
# Generate a unique job name
timestamp = datetime.now().strftime('%Y%m%d%H%M%S')
job_name = f"casl-transcription-{timestamp}"
s3_key = f"{S3_PREFIX}{job_name}{file_extension}"
# Upload to S3
logger.info(f"Uploading {file_name} to S3 bucket {S3_BUCKET}")
try:
with open(audio_path, 'rb') as audio_file:
s3_client.upload_fileobj(audio_file, S3_BUCKET, s3_key)
except Exception as e:
logger.error(f"Failed to upload to S3: {str(e)}")
# If upload fails, try to create the bucket
try:
s3_client.create_bucket(Bucket=S3_BUCKET)
logger.info(f"Created S3 bucket: {S3_BUCKET}")
# Try upload again
with open(audio_path, 'rb') as audio_file:
s3_client.upload_fileobj(audio_file, S3_BUCKET, s3_key)
except Exception as bucket_error:
logger.error(f"Failed to create bucket and upload: {str(bucket_error)}")
return "Error: Failed to upload audio file. Please check your AWS permissions."
# Start transcription job
logger.info(f"Starting transcription job: {job_name}")
media_format = file_extension.lower()[1:] # Remove the dot
if media_format == 'webm':
media_format = 'webm' # Amazon Transcribe expects this
# Determine language settings based on patient age
if patient_age < 10:
# For younger children, enabling child language model is helpful
language_options = {
'LanguageCode': 'en-US',
'Settings': {
'LanguageModelName': 'ChildLanguage'
}
}
else:
language_options = {
'LanguageCode': 'en-US'
}
transcribe_client.start_transcription_job(
TranscriptionJobName=job_name,
Media={
'MediaFileUri': f"s3://{S3_BUCKET}/{s3_key}"
},
MediaFormat=media_format,
**language_options,
Settings={
'ShowSpeakerLabels': True,
'MaxSpeakerLabels': 2 # Typically patient + clinician
}
)
# Wait for the job to complete (with timeout)
logger.info("Waiting for transcription to complete...")
max_tries = 30 # 5 minutes max wait
tries = 0
while tries < max_tries:
try:
job = transcribe_client.get_transcription_job(TranscriptionJobName=job_name)
status = job['TranscriptionJob']['TranscriptionJobStatus']
if status == 'COMPLETED':
# Get the transcript
transcript_uri = job['TranscriptionJob']['Transcript']['TranscriptFileUri']
# Download the transcript
import urllib.request
import json
with urllib.request.urlopen(transcript_uri) as response:
transcript_json = json.loads(response.read().decode('utf-8'))
# Convert to CHAT format
chat_transcript = format_as_chat(transcript_json)
return chat_transcript
elif status == 'FAILED':
reason = job['TranscriptionJob'].get('FailureReason', 'Unknown failure')
logger.error(f"Transcription job failed: {reason}")
return f"Error: Transcription failed - {reason}"
# Still in progress, wait and try again
tries += 1
time.sleep(10) # Check every 10 seconds
except Exception as e:
logger.error(f"Error checking transcription job: {str(e)}")
return f"Error getting transcription: {str(e)}"
# If we got here, we timed out
return "Error: Transcription timed out. The process is taking longer than expected."
except Exception as e:
logger.exception("Error in audio transcription")
return f"Error transcribing audio: {str(e)}"
def format_as_chat(transcript_json):
"""Format the Amazon Transcribe JSON result as CHAT format"""
try:
# Get transcript items
items = transcript_json['results']['items']
# Get speaker labels if available
speakers = {}
if 'speaker_labels' in transcript_json['results']:
speaker_segments = transcript_json['results']['speaker_labels']['segments']
# Map each item to its speaker
for segment in speaker_segments:
for item in segment['items']:
start_time = item['start_time']
speakers[start_time] = segment['speaker_label']
# Build transcript by combining words into utterances by speaker
current_speaker = None
current_utterance = []
utterances = []
for item in items:
# Skip non-pronunciation items (like punctuation)
if item['type'] != 'pronunciation':
continue
word = item['alternatives'][0]['content']
start_time = item.get('start_time')
# Determine speaker if available
speaker = speakers.get(start_time, 'spk_0')
# If speaker changed, start a new utterance
if speaker != current_speaker and current_utterance:
utterances.append((current_speaker, ' '.join(current_utterance)))
current_utterance = []
current_speaker = speaker
current_utterance.append(word)
# Add the last utterance
if current_utterance:
utterances.append((current_speaker, ' '.join(current_utterance)))
# Format as CHAT
chat_lines = []
for speaker, text in utterances:
# Map speakers to CHAT format
# Assuming spk_0 is the patient (PAR) and spk_1 is the clinician (INV)
chat_speaker = "*PAR:" if speaker == "spk_0" else "*INV:"
chat_lines.append(f"{chat_speaker} {text}.")
return '\n'.join(chat_lines)
except Exception as e:
logger.exception("Error formatting transcript")
return "*PAR: (Error formatting transcript)"
def generate_demo_transcription():
"""Generate a simulated transcription response"""
return """*PAR: today I want to tell you about my favorite toy.
*PAR: it's a &-um teddy bear that I got for my birthday.
*PAR: he has &-um brown fur and a red bow.
*PAR: I like to sleep with him every night.
*PAR: sometimes I take him to school in my backpack.
*INV: what's your teddy bear's name?
*PAR: his name is &-um Brownie because he's brown."""
def generate_demo_response(prompt):
"""Generate a response using Bedrock if available, otherwise return a demo response"""
# This function will attempt to call Bedrock, and only fall back to the demo response
# if Bedrock is not available or fails
# Try to call Bedrock first if client is available
if bedrock_client:
try:
return call_bedrock(prompt)
except Exception as e:
logger.error(f"Error calling Bedrock: {str(e)}")
logger.info("Falling back to demo response")
# Continue to fallback response if Bedrock call fails
# Fallback demo response
logger.warning("Using demo response - Bedrock client not available or call failed")
return """<SPEECH_FACTORS_START>
Difficulty producing fluent speech: 8, 65
Examples:
- "today I would &-um like to talk about &-um a fun trip I took last &-um summer with my family"
- "we went to the &-um &-um beach [//] no to the mountains [//] I mean the beach actually"
Word retrieval issues: 6, 72
Examples:
- "what do you call those &-um &-um sprinkles! that's the word"
- "sometimes I forget [//] forgetted [: forgot] [*] what they call those things we built"
Grammatical errors: 4, 58
Examples:
- "after swimming we [//] I eat [: ate] [*] &-um ice cream"
- "sometimes I forget [//] forgetted [: forgot] [*] what they call those things we built"
Repetitions and revisions: 5, 62
Examples:
- "we [/] we stayed for &-um three no [//] four days"
- "we went to the &-um &-um beach [//] no to the mountains [//] I mean the beach actually"
<SPEECH_FACTORS_END>
<CASL_SKILLS_START>
Lexical/Semantic Skills: Standard Score (92), Percentile Rank (30%), Average Performance
Examples:
- "what do you call those &-um &-um sprinkles! that's the word"
- "we went to the &-um &-um beach [//] no to the mountains [//] I mean the beach actually"
Syntactic Skills: Standard Score (87), Percentile Rank (19%), Low Average Performance
Examples:
- "my brother he [//] he helped me dig a big hole"
- "after swimming we [//] I eat [: ate] [*] &-um ice cream with &-um chocolate things on top"
Supralinguistic Skills: Standard Score (90), Percentile Rank (25%), Average Performance
Examples:
- "sometimes I wonder [/] wonder where fishies [: fish] [*] go when it's cold"
- "maybe they have [/] have houses under the water"
<CASL_SKILLS_END>
<TREATMENT_RECOMMENDATIONS_START>
- Implement word-finding strategies with semantic cuing focused on everyday objects and activities, using the patient's beach experience as a context (e.g., "sprinkles," "castles")
- Practice structured narrative tasks with visual supports to reduce revisions and improve sequencing
- Use sentence formulation exercises focusing on verb tense consistency (addressing errors like "forgetted" and "eat" for "ate")
- Incorporate self-monitoring techniques to help identify and correct grammatical errors
- Work on increasing vocabulary specificity (e.g., "things on top" to "sprinkles")
<TREATMENT_RECOMMENDATIONS_END>
<EXPLANATION_START>
This child demonstrates moderate word-finding difficulties with compensatory strategies including fillers ("&-um") and repetitions. The frequent use of self-corrections shows good metalinguistic awareness, but the pauses and repairs impact conversational fluency. Syntactic errors primarily involve verb tense inconsistency. Overall, the pattern suggests a mild-to-moderate language disorder with stronger receptive than expressive skills.
<EXPLANATION_END>
<ADDITIONAL_ANALYSIS_START>
The child shows relative strengths in maintaining topic coherence and conveying a complete narrative structure despite the language challenges. The pattern of errors suggests that word-finding difficulties and processing speed are primary concerns rather than conceptual or cognitive issues. Semantic network activities that strengthen word associations would likely be beneficial, particularly when paired with visual supports.
<ADDITIONAL_ANALYSIS_END>
<DIAGNOSTIC_IMPRESSIONS_START>
Based on the language sample, this child presents with a profile consistent with a mild-to-moderate expressive language disorder. The most prominent features include:
1. Word-finding difficulties characterized by fillers, pauses, and self-corrections when attempting to retrieve specific vocabulary
2. Grammatical challenges primarily affecting verb tense consistency and morphological markers
3. Relatively intact narrative structure and topic maintenance
These findings suggest intervention should focus on word retrieval strategies, grammatical form practice, and continued support for narrative development, with an emphasis on fluency and self-monitoring.
<DIAGNOSTIC_IMPRESSIONS_END>
<ERROR_EXAMPLES_START>
Word-finding difficulties:
- "what do you call those &-um &-um sprinkles! that's the word"
- "we went to the &-um &-um beach [//] no to the mountains [//] I mean the beach actually"
- "there was lots of &-um &-um swimming and &-um sun"
Grammatical errors:
- "after swimming we [//] I eat [: ate] [*] &-um ice cream"
- "sometimes I forget [//] forgetted [: forgot] [*] what they call those things we built"
- "we saw [/] saw fishies [: fish] [*] swimming in the water"
Repetitions and revisions:
- "we [/] we stayed for &-um three no [//] four days"
- "I want to go back to the beach [/] beach next year"
- "sometimes I wonder [/] wonder where fishies [: fish] [*] go when it's cold"
<ERROR_EXAMPLES_END>"""
def parse_casl_response(response):
"""Parse the LLM response for CASL analysis into structured data"""
# Extract speech factors section using section markers
speech_factors_section = ""
factors_pattern = re.compile(r"<SPEECH_FACTORS_START>(.*?)<SPEECH_FACTORS_END>", re.DOTALL)
factors_match = factors_pattern.search(response)
if factors_match:
speech_factors_section = factors_match.group(1).strip()
else:
speech_factors_section = "Error extracting speech factors from analysis."
# Extract CASL skills section
casl_section = ""
casl_pattern = re.compile(r"<CASL_SKILLS_START>(.*?)<CASL_SKILLS_END>", re.DOTALL)
casl_match = casl_pattern.search(response)
if casl_match:
casl_section = casl_match.group(1).strip()
else:
casl_section = "Error extracting CASL skills from analysis."
# Extract treatment recommendations
treatment_text = ""
treatment_pattern = re.compile(r"<TREATMENT_RECOMMENDATIONS_START>(.*?)<TREATMENT_RECOMMENDATIONS_END>", re.DOTALL)
treatment_match = treatment_pattern.search(response)
if treatment_match:
treatment_text = treatment_match.group(1).strip()
else:
treatment_text = "Error extracting treatment recommendations from analysis."
# Extract explanation section
explanation_text = ""
explanation_pattern = re.compile(r"<EXPLANATION_START>(.*?)<EXPLANATION_END>", re.DOTALL)
explanation_match = explanation_pattern.search(response)
if explanation_match:
explanation_text = explanation_match.group(1).strip()
else:
explanation_text = "Error extracting clinical explanation from analysis."
# Extract additional analysis
additional_analysis = ""
additional_pattern = re.compile(r"<ADDITIONAL_ANALYSIS_START>(.*?)<ADDITIONAL_ANALYSIS_END>", re.DOTALL)
additional_match = additional_pattern.search(response)
if additional_match:
additional_analysis = additional_match.group(1).strip()
# Extract diagnostic impressions
diagnostic_impressions = ""
diagnostic_pattern = re.compile(r"<DIAGNOSTIC_IMPRESSIONS_START>(.*?)<DIAGNOSTIC_IMPRESSIONS_END>", re.DOTALL)
diagnostic_match = diagnostic_pattern.search(response)
if diagnostic_match:
diagnostic_impressions = diagnostic_match.group(1).strip()
# Extract specific error examples
specific_errors_text = ""
errors_pattern = re.compile(r"<ERROR_EXAMPLES_START>(.*?)<ERROR_EXAMPLES_END>", re.DOTALL)
errors_match = errors_pattern.search(response)
if errors_match:
specific_errors_text = errors_match.group(1).strip()
# Create full report text
full_report = f"""
## Speech Factors Analysis
{speech_factors_section}
## CASL Skills Assessment
{casl_section}
## Treatment Recommendations
{treatment_text}
## Clinical Explanation
{explanation_text}
"""
if additional_analysis:
full_report += f"\n## Additional Analysis\n\n{additional_analysis}"
if diagnostic_impressions:
full_report += f"\n## Diagnostic Impressions\n\n{diagnostic_impressions}"
if specific_errors_text:
full_report += f"\n## Detailed Error Examples\n\n{specific_errors_text}"
return {
'speech_factors': speech_factors_section,
'casl_data': casl_section,
'treatment_suggestions': treatment_text,
'explanation': explanation_text,
'additional_analysis': additional_analysis,
'diagnostic_impressions': diagnostic_impressions,
'specific_errors': specific_errors_text,
'full_report': full_report,
'raw_response': response
}
def analyze_transcript(transcript, age, gender):
"""Analyze a speech transcript using Claude"""
# CASL-2 assessment cheat sheet
cheat_sheet = """
# Speech-Language Pathologist Analysis Cheat Sheet
## Types of Speech Patterns to Identify:
1. Difficulty producing fluent, grammatical speech
- Fillers (um, uh) and pauses
- False starts and revisions
- Incomplete sentences
2. Word retrieval issues
- Pauses before content words
- Circumlocutions (talking around a word)
- Word substitutions
3. Grammatical errors
- Verb tense inconsistencies
- Subject-verb agreement errors
- Morphological errors (plurals, possessives)
4. Repetitions and revisions
- Word or phrase repetitions [/]
- Self-corrections [//]
- Retracing
5. Neologisms
- Made-up words
- Word blends
6. Perseveration
- Inappropriate repetition of ideas
- Recurring themes
7. Comprehension issues
- Topic maintenance difficulties
- Non-sequiturs
- Inappropriate responses
"""
# Instructions for the analysis
instructions = """
Analyze this speech transcript to identify specific patterns and provide a detailed CASL-2 (Comprehensive Assessment of Spoken Language) assessment.
For each speech pattern you identify:
1. Count the occurrences in the transcript
2. Estimate a percentile (how typical/atypical this is for the age)
3. Provide DIRECT QUOTES from the transcript as evidence
Then assess the following CASL-2 domains:
1. Lexical/Semantic Skills:
- Assess vocabulary diversity, word-finding abilities, semantic precision
- Provide Standard Score (mean=100, SD=15), percentile rank, and performance level
- Include SPECIFIC QUOTES as evidence
2. Syntactic Skills:
- Evaluate grammatical accuracy, sentence complexity, morphological skills
- Provide Standard Score, percentile rank, and performance level
- Include SPECIFIC QUOTES as evidence
3. Supralinguistic Skills:
- Assess figurative language use, inferencing, and abstract reasoning
- Provide Standard Score, percentile rank, and performance level
- Include SPECIFIC QUOTES as evidence
YOUR RESPONSE MUST USE THESE EXACT SECTION MARKERS FOR PARSING:
<SPEECH_FACTORS_START>
Difficulty producing fluent, grammatical speech: (occurrences), (percentile)
Examples:
- "(direct quote from transcript)"
- "(direct quote from transcript)"
Word retrieval issues: (occurrences), (percentile)
Examples:
- "(direct quote from transcript)"
- "(direct quote from transcript)"
(And so on for each factor)
<SPEECH_FACTORS_END>
<CASL_SKILLS_START>
Lexical/Semantic Skills: Standard Score (X), Percentile Rank (X%), Performance Level
Examples:
- "(direct quote showing strength or weakness)"
- "(direct quote showing strength or weakness)"
Syntactic Skills: Standard Score (X), Percentile Rank (X%), Performance Level
Examples:
- "(direct quote showing strength or weakness)"
- "(direct quote showing strength or weakness)"
Supralinguistic Skills: Standard Score (X), Percentile Rank (X%), Performance Level
Examples:
- "(direct quote showing strength or weakness)"
- "(direct quote showing strength or weakness)"
<CASL_SKILLS_END>
<TREATMENT_RECOMMENDATIONS_START>
- (treatment recommendation)
- (treatment recommendation)
- (treatment recommendation)
<TREATMENT_RECOMMENDATIONS_END>
<EXPLANATION_START>
(brief diagnostic rationale based on findings)
<EXPLANATION_END>
<ADDITIONAL_ANALYSIS_START>
(specific insights that would be helpful for treatment planning)
<ADDITIONAL_ANALYSIS_END>
<DIAGNOSTIC_IMPRESSIONS_START>
(summarize findings across domains using specific examples and clear explanations)
<DIAGNOSTIC_IMPRESSIONS_END>
<ERROR_EXAMPLES_START>
(Copy all the specific quote examples here again, organized by error type or skill domain)
<ERROR_EXAMPLES_END>
MOST IMPORTANT:
1. Use EXACTLY the section markers provided (like <SPEECH_FACTORS_START>) to make parsing reliable
2. For EVERY factor and domain you analyze, you MUST provide direct quotes from the transcript as evidence
3. Be very specific and cite the exact text
4. Do not omit any of the required sections
"""
# Prepare prompt for Claude with the user's role context
role_context = """
You are a speech pathologist, a healthcare professional who specializes in evaluating, diagnosing, and treating communication disorders, including speech, language, cognitive-communication, voice, swallowing, and fluency disorders. Your role is to help patients improve their speech and communication skills through various therapeutic techniques and exercises.
You are working with a student with speech impediments.
The most important thing is that you stay kind to the child. Be constructive and helpful rather than critical.
"""
prompt = f"""
{role_context}
You are analyzing a transcript for a patient who is {age} years old and {gender}.
TRANSCRIPT:
{transcript}
{cheat_sheet}
{instructions}
Remember to be precise but compassionate in your analysis. Use direct quotes from the transcript for every factor and domain you analyze.
"""
# Call the appropriate API or fallback to demo mode
if bedrock_client:
response = call_bedrock(prompt)
else:
response = generate_demo_response(prompt)
# Parse the response
results = parse_casl_response(response)
return results
def export_pdf(results, patient_name="", record_id="", age="", gender="", assessment_date="", clinician=""):
"""Export analysis results to a PDF report"""
global DOWNLOADS_DIR
# Check if ReportLab is available
if not REPORTLAB_AVAILABLE:
return "ERROR: PDF export is not available - ReportLab library is not installed. Please run 'pip install reportlab'."
try:
# Generate a safe filename
if patient_name:
safe_name = f"{patient_name.replace(' ', '_')}"
else:
safe_name = f"speech_analysis_{datetime.now().strftime('%Y%m%d%H%M%S')}"
# Make sure the downloads directory exists
try:
os.makedirs(DOWNLOADS_DIR, exist_ok=True)
except Exception as e:
logger.warning(f"Could not access downloads directory: {str(e)}")
# Fallback to temp directory
DOWNLOADS_DIR = os.path.join(tempfile.gettempdir(), "casl_downloads")
os.makedirs(DOWNLOADS_DIR, exist_ok=True)
# Create the PDF path in our downloads directory
pdf_path = os.path.join(DOWNLOADS_DIR, f"{safe_name}.pdf")
# Create the PDF document
doc = SimpleDocTemplate(pdf_path, pagesize=letter)
styles = getSampleStyleSheet()
# Create enhanced custom styles
styles.add(ParagraphStyle(
name='Heading1',
parent=styles['Heading1'],
fontSize=16,
spaceAfter=12,
textColor=colors.navy
))
styles.add(ParagraphStyle(
name='Heading2',
parent=styles['Heading2'],
fontSize=14,
spaceAfter=10,
spaceBefore=10,
textColor=colors.darkblue
))
styles.add(ParagraphStyle(
name='Heading3',
parent=styles['Heading2'],
fontSize=12,
spaceAfter=8,
spaceBefore=8,
textColor=colors.darkblue
))
styles.add(ParagraphStyle(
name='BodyText',
parent=styles['BodyText'],
fontSize=11,
spaceAfter=8,
leading=14
))
styles.add(ParagraphStyle(
name='BulletPoint',
parent=styles['BodyText'],
fontSize=11,
leftIndent=20,
firstLineIndent=-15,
spaceAfter=4,
leading=14
))
# Convert markdown to PDF elements
story = []
# Add title and date
story.append(Paragraph("Speech Language Assessment Report", styles['Title']))
story.append(Spacer(1, 12))
# Add patient information table
if patient_name or record_id or age or gender:
# Prepare patient info data
data = []
if patient_name:
data.append(["Patient Name:", patient_name])
if record_id:
data.append(["Record ID:", record_id])
if age:
data.append(["Age:", f"{age} years"])
if gender:
data.append(["Gender:", gender])
if assessment_date:
data.append(["Assessment Date:", assessment_date])
if clinician:
data.append(["Clinician:", clinician])
if data:
# Create a table with the data
patient_table = Table(data, colWidths=[120, 350])
patient_table.setStyle(TableStyle([
('BACKGROUND', (0, 0), (0, -1), colors.lightgrey),
('TEXTCOLOR', (0, 0), (0, -1), colors.darkblue),
('ALIGN', (0, 0), (0, -1), 'RIGHT'),
('ALIGN', (1, 0), (1, -1), 'LEFT'),
('FONTNAME', (0, 0), (0, -1), 'Helvetica-Bold'),
('BOTTOMPADDING', (0, 0), (-1, -1), 6),
('TOPPADDING', (0, 0), (-1, -1), 6),
('GRID', (0, 0), (-1, -1), 0.5, colors.lightgrey),
]))
story.append(patient_table)
story.append(Spacer(1, 12))
# Add clinical analysis sections
story.append(Paragraph("Speech Factors Analysis", styles['Heading1']))
speech_factors_paragraphs = []
for line in results['speech_factors'].split('\n'):
line = line.strip()
if not line:
continue
if line.startswith('- '):
story.append(Paragraph(f"• {line[2:]}", styles['BulletPoint']))
else:
story.append(Paragraph(line, styles['BodyText']))
story.append(Spacer(1, 12))
story.append(Paragraph("CASL Skills Assessment", styles['Heading1']))
for line in results['casl_data'].split('\n'):
line = line.strip()
if not line:
continue
if line.startswith('- '):
story.append(Paragraph(f"• {line[2:]}", styles['BulletPoint']))
else:
story.append(Paragraph(line, styles['BodyText']))
story.append(Spacer(1, 12))
story.append(Paragraph("Treatment Recommendations", styles['Heading1']))
# Process treatment recommendations as bullet points
for line in results['treatment_suggestions'].split('\n'):
line = line.strip()
if not line:
continue
if line.startswith('- '):
story.append(Paragraph(f"• {line[2:]}", styles['BulletPoint']))
else:
story.append(Paragraph(line, styles['BodyText']))
story.append(Spacer(1, 12))
story.append(Paragraph("Clinical Explanation", styles['Heading1']))
story.append(Paragraph(results['explanation'], styles['BodyText']))
story.append(Spacer(1, 12))
if results['additional_analysis']:
story.append(Paragraph("Additional Analysis", styles['Heading1']))
story.append(Paragraph(results['additional_analysis'], styles['BodyText']))
story.append(Spacer(1, 12))
if results['diagnostic_impressions']:
story.append(Paragraph("Diagnostic Impressions", styles['Heading1']))
story.append(Paragraph(results['diagnostic_impressions'], styles['BodyText']))
story.append(Spacer(1, 12))
# Add footer with date
footer_text = f"Generated on: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}"
story.append(Spacer(1, 20))
story.append(Paragraph(footer_text, ParagraphStyle(
name='Footer',
parent=styles['Normal'],
fontSize=8,
textColor=colors.grey
)))
# Build the PDF
doc.build(story)
logger.info(f"Report saved as PDF: {pdf_path}")
return pdf_path
except Exception as e:
logger.exception("Error creating PDF")
return f"Error creating PDF: {str(e)}"
def create_interface():
"""Create the Gradio interface"""
# Set a theme compatible with Hugging Face Spaces
theme = gr.themes.Soft(
primary_hue="blue",
secondary_hue="indigo",
)
with gr.Blocks(title="Simple CASL Analysis Tool", theme=theme) as app:
gr.Markdown("# CASL Analysis Tool")
gr.Markdown("A simplified tool for analyzing speech transcripts and audio using CASL framework")
with gr.Tabs() as main_tabs:
# Analysis Tab
with gr.TabItem("Analysis", id=0):
with gr.Row():
with gr.Column(scale=1):
# Patient info
gr.Markdown("### Patient Information")
patient_name = gr.Textbox(label="Patient Name", placeholder="Enter patient name")
record_id = gr.Textbox(label="Record ID", placeholder="Enter record ID")
with gr.Row():
age = gr.Number(label="Age", value=8, minimum=1, maximum=120)
gender = gr.Radio(["male", "female", "other"], label="Gender", value="male")
assessment_date = gr.Textbox(
label="Assessment Date",
placeholder="MM/DD/YYYY",
value=datetime.now().strftime('%m/%d/%Y')
)
clinician_name = gr.Textbox(label="Clinician", placeholder="Enter clinician name")
# Transcript input
gr.Markdown("### Transcript")
sample_btn = gr.Button("Load Sample Transcript")
file_upload = gr.File(label="Upload transcript file (.txt or .cha)")
transcript = gr.Textbox(
label="Speech transcript (CHAT format preferred)",
placeholder="Enter transcript text or upload a file...",
lines=10
)
# Analysis button
analyze_btn = gr.Button("Analyze Transcript", variant="primary")
with gr.Column(scale=1):
# Results display
gr.Markdown("### Analysis Results")
analysis_output = gr.Markdown(label="Full Analysis")
# PDF export (only shown if ReportLab is available)
export_status = gr.Markdown("")
if REPORTLAB_AVAILABLE:
export_btn = gr.Button("Export as PDF", variant="secondary")
else:
gr.Markdown("⚠️ PDF export is disabled - ReportLab library is not installed")
# Transcription Tab
with gr.TabItem("Transcription", id=1):
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("### Audio Transcription")
gr.Markdown("Upload an audio recording to automatically transcribe it in CHAT format")
# Patient's age helps with transcription accuracy
transcription_age = gr.Number(label="Patient Age", value=8, minimum=1, maximum=120,
info="For children under 10, special language models may be used")
# Audio input
audio_input = gr.Audio(type="filepath", label="Upload Audio Recording",
elem_id="audio-input")
# Transcribe button
transcribe_btn = gr.Button("Transcribe Audio", variant="primary")
with gr.Column(scale=1):
# Transcription output
transcription_output = gr.Textbox(
label="Transcription Result",
placeholder="Transcription will appear here...",
lines=12
)
with gr.Row():
# Button to use transcription in analysis
copy_to_analysis_btn = gr.Button("Use for Analysis", variant="secondary")
# Status/info message
transcription_status = gr.Markdown("")
# Load sample transcript button
def load_sample():
return SAMPLE_TRANSCRIPT
sample_btn.click(load_sample, outputs=[transcript])
# File upload handler
file_upload.upload(process_upload, file_upload, transcript)
# Analysis button handler
def on_analyze_click(transcript_text, age_val, gender_val, patient_name_val, record_id_val, clinician_val, assessment_date_val):
if not transcript_text or len(transcript_text.strip()) < 50:
return "Error: Please provide a longer transcript for analysis."
try:
# Get the analysis results
results = analyze_transcript(transcript_text, age_val, gender_val)
# Return the full report
return results['full_report']
except Exception as e:
logger.exception("Error during analysis")
return f"Error during analysis: {str(e)}"
analyze_btn.click(
on_analyze_click,
inputs=[
transcript, age, gender,
patient_name, record_id, clinician_name, assessment_date
],
outputs=[analysis_output]
)
# PDF export function
def on_export_pdf(report_text, p_name, p_record_id, p_age, p_gender, p_date, p_clinician):
# Check if ReportLab is available
if not REPORTLAB_AVAILABLE:
return "ERROR: PDF export is not available because the ReportLab library is not installed. Please install it with 'pip install reportlab'."
if not report_text or len(report_text.strip()) < 50:
return "Error: Please run the analysis first before exporting to PDF."
try:
# Parse the report text back into sections
results = {
'speech_factors': '',
'casl_data': '',
'treatment_suggestions': '',
'explanation': '',
'additional_analysis': '',
'diagnostic_impressions': '',
}
sections = report_text.split('##')
for section in sections:
section = section.strip()
if not section:
continue
title_content = section.split('\n', 1)
if len(title_content) < 2:
continue
title = title_content[0].strip()
content = title_content[1].strip()
if "Speech Factors Analysis" in title:
results['speech_factors'] = content
elif "CASL Skills Assessment" in title:
results['casl_data'] = content
elif "Treatment Recommendations" in title:
results['treatment_suggestions'] = content
elif "Clinical Explanation" in title:
results['explanation'] = content
elif "Additional Analysis" in title:
results['additional_analysis'] = content
elif "Diagnostic Impressions" in title:
results['diagnostic_impressions'] = content
pdf_path = export_pdf(
results,
patient_name=p_name,
record_id=p_record_id,
age=p_age,
gender=p_gender,
assessment_date=p_date,
clinician=p_clinician
)
# Check if the export was successful
if pdf_path.startswith("ERROR:"):
return pdf_path
# Make it downloadable in Hugging Face Spaces
download_link = f'<a href="file={pdf_path}" download="{os.path.basename(pdf_path)}">Download PDF Report</a>'
return f"Report saved as PDF: {pdf_path}<br>{download_link}"
except Exception as e:
logger.exception("Error exporting to PDF")
return f"Error creating PDF: {str(e)}"
# Only set up the PDF export button if ReportLab is available
if REPORTLAB_AVAILABLE:
export_btn.click(
on_export_pdf,
inputs=[
analysis_output,
patient_name,
record_id,
age,
gender,
assessment_date,
clinician_name
],
outputs=[export_status]
)
# Transcription button handler
def on_transcribe_audio(audio_path, age_val):
try:
if not audio_path:
return "Please upload an audio file to transcribe.", "Error: No audio file provided."
# Process the audio file with Amazon Transcribe
transcription = transcribe_audio(audio_path, age_val)
# Return status message based on whether it's a demo or real transcription
if not transcribe_client:
status_msg = "⚠️ Demo mode: Using example transcription (AWS credentials not configured)"
else:
status_msg = "✅ Transcription completed successfully"
return transcription, status_msg
except Exception as e:
logger.exception("Error transcribing audio")
return f"Error: {str(e)}", f"❌ Transcription failed: {str(e)}"
# Connect the transcribe button to its handler
transcribe_btn.click(
on_transcribe_audio,
inputs=[audio_input, transcription_age],
outputs=[transcription_output, transcription_status]
)
# Copy transcription to analysis tab
def copy_to_analysis(transcription):
return transcription, gr.update(selected=0) # Switch to Analysis tab
copy_to_analysis_btn.click(
copy_to_analysis,
inputs=[transcription_output],
outputs=[transcript, main_tabs]
)
return app
# Create requirements.txt file for HuggingFace Spaces
def create_requirements_file():
requirements = [
"gradio>=4.0.0",
"pandas",
"numpy",
"Pillow",
"boto3>=1.28.0", # Required for AWS services
"botocore>=1.31.0", # Required for AWS services
"reportlab>=3.6.0" # Optional for PDF exports
]
with open("requirements.txt", "w") as f:
for req in requirements:
f.write(f"{req}\n")
if __name__ == "__main__":
# Create requirements.txt for HuggingFace Spaces
create_requirements_file()
# Check for AWS credentials
if not AWS_ACCESS_KEY or not AWS_SECRET_KEY:
print("NOTE: AWS credentials not found. The app will run in demo mode with simulated responses.")
print("To enable full functionality, set AWS_ACCESS_KEY and AWS_SECRET_KEY environment variables.")
app = create_interface()
app.launch(show_api=False) # Disable API tab for security