stt.py

# stt.py
from faster_whisper import WhisperModel

# Load model once
stt_model = WhisperModel(
    "small",
    device="cpu",        # change to "cuda" if GPU available
    compute_type="int8"
)

def speech_to_text(audio_path):
    segments, info = stt_model.transcribe(audio_path)
    text = ""
    for segment in segments:
        text += segment.text
    return text