examples/make_callbot_test_media.py

#!/usr/bin/python3
# -*- coding: utf-8 -*-
import argparse
from collections import defaultdict
import json
from pathlib import Path

import numpy as np
from scipy.io import wavfile
import torch
from tqdm import tqdm
from typing import List

from project_settings import project_path


def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--call_monitor_examples_wav_dir",
        default=(project_path / "data/call_monitor_examples_wav/id-ID").as_posix(),
        type=str
    )
    parser.add_argument(
        "--output_dir",
        default=(project_path / "data/voice_test_examples").as_posix(),
        type=str
    )
    parser.add_argument("--n_samples", default=7, type=int)
    args = parser.parse_args()
    return args


def save_media(output_dir: Path, language: str, category: str, call_id: str,
               early_media: np.ndarray, active_media: np.ndarray = None,
               sample_rate: int = 8000):
    early_media_filename = output_dir / "{}/{}/early_media_{}.wav".format(language, category, call_id)
    active_media_filename = output_dir / "{}/{}/active_media_{}.wav".format(language, category, call_id)

    early_media_filename.parent.mkdir(parents=True, exist_ok=True)
    wavfile.write(early_media_filename.as_posix(), sample_rate, early_media)

    if active_media is not None:
        active_media_filename.parent.mkdir(parents=True, exist_ok=True)
        wavfile.write(active_media_filename.as_posix(), sample_rate, active_media)


def main():
    args = get_args()

    call_monitor_examples_wav_dir = Path(args.call_monitor_examples_wav_dir)
    output_dir = Path(args.output_dir)

    counter = defaultdict(int)

    metadata_file = call_monitor_examples_wav_dir / "metadata.json"
    with open(metadata_file.as_posix(), "r", encoding="utf-8") as f:
        metadata = json.load(f)

    for meta in metadata:
        filename = meta["filename"]
        early_media_label = meta["early_media_label"]
        early_media_ts = meta["early_media_ts"]
        on_answer_label = meta["on_answer_label"]
        on_answer_ts = meta["on_answer_ts"]

        filename = call_monitor_examples_wav_dir / filename

        call_id = filename.stem
        language = filename.parts[-2]

        sample_rate, signal = wavfile.read(filename.as_posix())

        if on_answer_ts is None:
            early_media = signal
            active_media = None
        else:
            early_media_n_samples = int(on_answer_ts / 1000 * sample_rate)
            early_media = signal[:early_media_n_samples]
            active_media = signal[early_media_n_samples:]

            append_length = 16000 * 15
            if len(active_media) < 16000:
                multiple = append_length / len(active_media)
                active_media_ = [active_media] + [active_media[-16000:]] * int(multiple)
            else:
                active_media_ = [active_media] + [active_media[-16000:]] * 15

            active_media = np.concatenate(active_media_, axis=0)

        category = "other"
        if on_answer_label in ("voicemail",):
            category = "01"
            if counter[category] > args.n_samples:
                continue
            counter[category] += 1
            save_media(
                output_dir=output_dir,
                language=language,
                category=category,
                call_id=call_id,
                early_media=early_media,
                active_media=active_media,
                sample_rate=sample_rate,
            )

        if on_answer_label in ("mute", "white_noise"):
            category = "03"
            if counter[category] > args.n_samples:
                continue
            counter[category] += 1
            save_media(
                output_dir=output_dir,
                language=language,
                category=category,
                call_id=call_id,
                early_media=early_media,
                active_media=active_media,
                sample_rate=sample_rate,
            )

        if early_media_label is not None and on_answer_label in ("voicemail",):
            category = "04"
            if counter[category] > args.n_samples:
                continue
            counter[category] += 1
            save_media(
                output_dir=output_dir,
                language=language,
                category=category,
                call_id=call_id,
                early_media=early_media,
                active_media=active_media,
                sample_rate=sample_rate,
            )

        if on_answer_label in ("voice",):
            category = "05"
            if counter[category] > args.n_samples:
                continue
            counter[category] += 1
            save_media(
                output_dir=output_dir,
                language=language,
                category=category,
                call_id=call_id,
                early_media=early_media,
                active_media=active_media,
                sample_rate=sample_rate,
            )

        if on_answer_ts > 25000:
            category = "06"
            if counter[category] > args.n_samples:
                continue
            counter[category] += 1
            save_media(
                output_dir=output_dir,
                language=language,
                category=category,
                call_id=call_id,
                early_media=early_media,
                active_media=active_media,
                sample_rate=sample_rate,
            )

        if category == "other":
            if counter[category] > args.n_samples:
                continue
            counter[category] += 1
            save_media(
                output_dir=output_dir,
                language=language,
                category=category,
                call_id=call_id,
                early_media=early_media,
                active_media=active_media,
                sample_rate=sample_rate,
            )

    return


if __name__ == '__main__':
    main()