An audio preprocessor module for applications that uses Whisper or Faster-Whisper to STT.

audio_preprocessor.py

import numpy as np
import noisereduce as nr
from loguru import logger
from pydub import AudioSegment
from pydub.silence import split_on_silence

def load_audio_file(audio_path: str) -> AudioSegment | None:
    try:
        audio = AudioSegment\
            .from_file(audio_path)\
            .set_channels(1)
        return audio
    except Exception as e:
        logger.error(f"Failed to load the audio file: {e}")
        return None

def reduce_noise(audio: AudioSegment) -> AudioSegment:
    # resolves audio data into the format that noisereduce can handle
    samplerate = audio.frame_rate
    samples = np.array(audio.get_array_of_samples()).astype(np.float32)
    reduced_noise_samples = nr.reduce_noise(y=samples, sr=samplerate, prop_decrease=0.8)

    processed_audio = AudioSegment(
        reduced_noise_samples.astype(np.int16).tobytes(),
        frame_rate=samplerate,
        sample_width=2,  # 16-bit audio
        channels=1
    )
    return processed_audio

def calculate_silence_threshold(audio: AudioSegment) -> float:
    candidate_thresh = audio.dBFS - 8
    safe_floor_thresh = -50.0
    final_thresh = min(candidate_thresh, safe_floor_thresh)
    return final_thresh

def audio_preprocess(audio_path: str) -> bool:
    audio = load_audio_file(audio_path)
    if audio is None:
        return False

    processed_audio = reduce_noise(audio)
    silence_thresh = calculate_silence_threshold(processed_audio)

    chunks = split_on_silence(
        processed_audio,
        min_silence_len=500,
        silence_thresh=int(silence_thresh),
        keep_silence=150,
    )

    # merge the splited audio data
    if not chunks:
        logger.warning("Failed detect the voice segment. will use the original audio file.")
        return False
    else:
        final_audio = AudioSegment.empty()
        for chunk in chunks:
            final_audio += chunk

    try:
        final_audio.export(audio_path, format="wav")
        return True
    except Exception as e:
        logger.error(f"Failed to export the processed audio data: {e}")
        return False