init
This commit is contained in:
mike
@@ -15,11 +15,13 @@ import json
|
||||
from datetime import datetime
|
||||
from concurrent.futures import ThreadPoolExecutor, as_completed
|
||||
|
||||
# Choose your Whisper backend here:
|
||||
# For faster-whisper (recommended):
|
||||
# Whisper transcription (using faster-whisper for optimal performance)
|
||||
from faster_whisper import WhisperModel
|
||||
|
||||
# LLM integration
|
||||
# Sentence extraction for stitching chunks
|
||||
from sentence_extractor import SentenceExtractor, SentenceCleaner
|
||||
|
||||
# LLM integration (optional)
|
||||
try:
|
||||
import ollama
|
||||
OLLAMA_AVAILABLE = True
|
||||
@@ -27,10 +29,6 @@ except ImportError:
|
||||
OLLAMA_AVAILABLE = False
|
||||
|
||||
|
||||
# # For regular whisper (comment out the line above and uncomment these):
|
||||
# import whisper
|
||||
|
||||
|
||||
class WindowsLoopbackAudioCapture:
|
||||
"""Capture Windows speaker output using WASAPI loopback"""
|
||||
|
||||
@@ -170,12 +168,6 @@ class WhisperStreamTranscriber:
|
||||
self.audio_buffer = np.array([], dtype=np.float32)
|
||||
self.lock = threading.Lock()
|
||||
|
||||
# # REGULAR WHISPER:
|
||||
# self.model = whisper.load_model(model_name)
|
||||
# self.language = language
|
||||
# self.audio_buffer = np.array([], dtype=np.float32)
|
||||
# self.lock = threading.Lock()
|
||||
|
||||
def add_audio(self, audio_chunk):
|
||||
"""Add new audio data to buffer"""
|
||||
with self.lock:
|
||||
@@ -222,19 +214,6 @@ class WhisperStreamTranscriber:
|
||||
print(f"❌ Transcription error: {e}")
|
||||
return None
|
||||
|
||||
# # REGULAR WHISPER:
|
||||
# try:
|
||||
# result = self.model.transcribe(
|
||||
# audio_to_process,
|
||||
# language=self.language,
|
||||
# task="transcribe",
|
||||
# fp16=False
|
||||
# )
|
||||
# return result["text"].strip()
|
||||
# except Exception as e:
|
||||
# print(f"❌ Transcription error: {e}")
|
||||
# return None
|
||||
|
||||
|
||||
class LocalLLMAnalyzer:
|
||||
"""Local LLM for fact-checking and question generation using Ollama"""
|
||||
@@ -536,6 +515,8 @@ Examples:
|
||||
help="Ollama model to use for LLM analysis (default: gpt-oss:20b)")
|
||||
parser.add_argument("--llm-debug", action="store_true",
|
||||
help="Show LLM raw responses for debugging")
|
||||
parser.add_argument("--sentence-mode", action="store_true",
|
||||
help="Extract complete sentences by stitching chunks together")
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
@@ -549,6 +530,8 @@ Examples:
|
||||
print(f"Output: {args.output}")
|
||||
if args.enable_llm:
|
||||
print(f"LLM Analysis: Enabled ({args.llm_model})")
|
||||
if args.sentence_mode:
|
||||
print(f"Sentence Mode: Enabled (stitching chunks into complete sentences)")
|
||||
|
||||
# Initialize audio capture
|
||||
try:
|
||||
@@ -591,6 +574,14 @@ Examples:
|
||||
print("Continuing without LLM analysis...")
|
||||
llm_analyzer = None
|
||||
|
||||
# Initialize sentence extractor (optional)
|
||||
sentence_extractor = None
|
||||
sentence_cleaner = None
|
||||
if args.sentence_mode:
|
||||
sentence_extractor = SentenceExtractor(max_buffer_words=150)
|
||||
sentence_cleaner = SentenceCleaner()
|
||||
print("✓ Sentence extraction initialized")
|
||||
|
||||
# Main processing loop
|
||||
print(f"\n✅ Started transcription. Press Ctrl+C to stop.\n{'=' * 50}")
|
||||
last_process_time = time.time()
|
||||
@@ -620,11 +611,45 @@ Examples:
|
||||
segment_count += 1
|
||||
timestamp = datetime.now().strftime("%H:%M:%S")
|
||||
|
||||
# Display transcription immediately (don't wait for LLM)
|
||||
print(f"[{timestamp}] {text}")
|
||||
# Sentence extraction mode
|
||||
if sentence_extractor:
|
||||
# Add chunk to extractor and get complete sentences
|
||||
sentences = sentence_extractor.add_chunk(text)
|
||||
|
||||
# LLM Analysis (run concurrently in background)
|
||||
if llm_analyzer:
|
||||
for sentence in sentences:
|
||||
# Clean the sentence
|
||||
cleaned = sentence_cleaner.clean(sentence) if sentence_cleaner else sentence
|
||||
if cleaned:
|
||||
print(f"[{timestamp}] 📝 {cleaned}")
|
||||
|
||||
# Save individual sentences
|
||||
if args.output and not llm_analyzer:
|
||||
save_transcript(cleaned, timestamp, args.output)
|
||||
|
||||
# LLM analysis on complete sentences
|
||||
if llm_analyzer:
|
||||
context = f"Sentence from segment {segment_count}"
|
||||
|
||||
def run_llm_analysis(txt, ctx, ts, seg_num):
|
||||
fc = llm_analyzer.fact_check(txt, ctx)
|
||||
qs = llm_analyzer.generate_augmenting_questions(txt, ctx)
|
||||
return {
|
||||
'timestamp': ts,
|
||||
'text': txt,
|
||||
'segment_count': seg_num,
|
||||
'fact_check': fc,
|
||||
'questions': qs
|
||||
}
|
||||
|
||||
future = llm_executor.submit(run_llm_analysis, cleaned, context, timestamp, segment_count)
|
||||
pending_llm_tasks[segment_count] = future
|
||||
else:
|
||||
# Standard mode: display chunks as-is
|
||||
# Display transcription immediately (don't wait for LLM)
|
||||
print(f"[{timestamp}] {text}")
|
||||
|
||||
# LLM Analysis (run concurrently in background) - only in non-sentence mode
|
||||
if llm_analyzer and not sentence_extractor:
|
||||
context = f"Segment {segment_count}"
|
||||
|
||||
# Submit LLM tasks to thread pool
|
||||
@@ -701,6 +726,34 @@ Examples:
|
||||
# Cleanup
|
||||
capturer.close()
|
||||
|
||||
# Flush sentence buffer if in sentence mode
|
||||
if sentence_extractor:
|
||||
print("\n📝 Flushing sentence buffer...")
|
||||
final_sentences = sentence_extractor.flush()
|
||||
for sentence in final_sentences:
|
||||
cleaned = sentence_cleaner.clean(sentence) if sentence_cleaner else sentence
|
||||
if cleaned:
|
||||
timestamp = datetime.now().strftime("%H:%M:%S")
|
||||
print(f"[{timestamp}] 📝 {cleaned}")
|
||||
|
||||
if args.output and not llm_analyzer:
|
||||
save_transcript(cleaned, timestamp, args.output)
|
||||
|
||||
# LLM analysis for flushed sentences
|
||||
if llm_analyzer:
|
||||
fact_check = llm_analyzer.fact_check(cleaned, "Final sentence")
|
||||
questions = llm_analyzer.generate_augmenting_questions(cleaned)
|
||||
display_enriched_output(cleaned, timestamp, fact_check, questions)
|
||||
|
||||
if args.output:
|
||||
data = {
|
||||
'timestamp': timestamp,
|
||||
'text': cleaned,
|
||||
'fact_check': fact_check,
|
||||
'questions': questions
|
||||
}
|
||||
save_enriched_transcript(data, args.output)
|
||||
|
||||
# Process remaining audio
|
||||
print("\nProcessing remaining audio...")
|
||||
final_text = transcriber.transcribe_chunk(min_duration=0)
|
||||
|
||||
Reference in New Issue
Block a user