AudioToTextNotes.py

import sounddevice as sd
import numpy as np
import whisper
import queue
import threading
import time
from transformers import pipeline
from datetime import datetime

# Load Whisper model on CPU
model = whisper.load_model("base")  # or "medium" for better accuracy

# Load summarization pipeline (CPU only)
summarizer = pipeline(
    "summarization",
    model="facebook/bart-large-cnn",
    device=-1
)

# Audio settings
SAMPLE_RATE = 16000
BLOCK_DURATION = 150  # x seconds
CHUNK_SIZE = int(SAMPLE_RATE * BLOCK_DURATION)

audio_queue = queue.Queue()

# Function to get the current time stamp
def get_current_timestamp():
    return datetime.now().strftime("%Y-%m-%d %H:%M:%S")

def audio_callback(indata, frames, time_info, status):
    audio_queue.put(indata.copy())

def record_audio():
    with sd.InputStream(samplerate=SAMPLE_RATE, channels=1, dtype="float32", callback=audio_callback):
        print("🎙️ Recording... (Press Ctrl+C to stop)")
        while True:
            time.sleep(1)

def save_raw_transcript(text: str):
    timestamp = get_current_timestamp()
    with open("raw_transcripts.txt", "a", encoding="utf-8") as f:
        f.write(f"{timestamp} - {text.strip()}\n\n")

def save_summary_to_file(summary: str):
    timestamp = get_current_timestamp()
    with open("notes.txt", "a", encoding="utf-8") as f:
        f.write(f"{timestamp} - {summary.strip()}\n\n")

def transcribe_and_summarize():
    buffer = []
    while True:
        try:
            data = audio_queue.get(timeout=10)
            buffer.append(data)

            if len(buffer) * data.shape[0] >= CHUNK_SIZE:
                audio_chunk = np.concatenate(buffer, axis=0)
                buffer = []

                print("📝 Transcribing chunk...")
                audio_np = audio_chunk.flatten()
                result = model.transcribe(audio_np, language="en")
                transcription = result["text"].strip()

                if len(transcription) > 20:
                    print("📄 Transcript:", transcription[:200] + "..." if len(transcription) > 200 else transcription)
                    save_raw_transcript(transcription)

                    print("✍️ Summarizing...")
                    summary = summarizer(
                        transcription,
                        max_length=150,  # 25*x sentences
                        min_length=75,  # 25*x sentences
                        do_sample=False
                    )[0]["summary_text"]

                    print("✅ Summary:\n", summary, "\n")
                    save_summary_to_file(summary)
        except queue.Empty:
            continue

# Start threads
record_thread = threading.Thread(target=record_audio, daemon=True)
process_thread = threading.Thread(target=transcribe_and_summarize, daemon=True)

record_thread.start()
process_thread.start()

# Keep the program alive
try:
    while True:
        time.sleep(10)
except KeyboardInterrupt:
    print("🛑 Exiting...")