pitch_analysis.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "authorship_tag": "ABX9TyMrHzjxYenywItbm/j399ze",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/kth0522/AI_news/blob/main/pitch_analysis.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "i2_QjDHlXKVj"
      },
      "outputs": [],
      "source": [
        "import os\n",
        "import torchaudio\n",
        "import torch\n",
        "import pandas as pd\n",
        "from tqdm import tqdm"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def analyze_folder_and_save_results(folder_path, output_folder_path, window_length=3.0, hop_length=None):\n",
        "    # Check if output folder exists, create if not\n",
        "    if not os.path.exists(output_folder_path):\n",
        "        os.makedirs(output_folder_path)\n",
        "\n",
        "    # Get list of all .wav files in the folder\n",
        "    wav_files = [f for f in os.listdir(folder_path) if f.lower().endswith('.wav')]\n",
        "\n",
        "    # Iterate over all WAV files in the folder with progress bar\n",
        "    for file_name in tqdm(wav_files, desc=\"Analyzing WAV files\"):\n",
        "        file_path = os.path.join(folder_path, file_name)\n",
        "\n",
        "        # Load the .wav file\n",
        "        waveform, sample_rate = torchaudio.load(file_path)\n",
        "\n",
        "        # Determine the window and hop length in samples\n",
        "        window_length_samples = int(window_length * sample_rate)\n",
        "        if hop_length is None:\n",
        "            hop_length = window_length_samples\n",
        "\n",
        "        # Compute the Short-Time Fourier Transform (STFT)\n",
        "        stft = torch.stft(\n",
        "            waveform,\n",
        "            n_fft=window_length_samples,\n",
        "            hop_length=hop_length,\n",
        "            win_length=window_length_samples,\n",
        "            window=torch.hann_window(window_length_samples),\n",
        "            return_complex=True\n",
        "        )\n",
        "\n",
        "        # Analyze each window\n",
        "        analysis_results = []\n",
        "        for i in range(stft.size(-1)):\n",
        "            window_stft = stft[:,:,i].abs()  # Get magnitude\n",
        "\n",
        "            mean = window_stft.mean().item()\n",
        "            var = window_stft.var().item()\n",
        "            max_val = window_stft.max().item()\n",
        "            min_val = window_stft.min().item()\n",
        "            median_val = window_stft.median().item()\n",
        "\n",
        "            analysis_results.append({\n",
        "                \"window\": i,\n",
        "                \"mean\": mean,\n",
        "                \"variance\": var,\n",
        "                \"max\": max_val,\n",
        "                \"min\": min_val,\n",
        "                \"median\": median_val\n",
        "            })\n",
        "\n",
        "        # Extract the number from the file name and create CSV file name\n",
        "        file_number = file_name.split('_')[-1].split('.')[0]\n",
        "        csv_file_name = f\"{file_number}.csv\"\n",
        "        output_csv_path = os.path.join(output_folder_path, csv_file_name)\n",
        "\n",
        "        # Convert results to a DataFrame and save to a CSV file\n",
        "        df = pd.DataFrame(analysis_results)\n",
        "        df.to_csv(output_csv_path, index=False)\n",
        "\n",
        "    return output_folder_path"
      ],
      "metadata": {
        "id": "YM4Oab0_XTkc"
      },
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "BASE_DIR = '/datasets/skku_anchor/'\n",
        "DATASET = 'mbn_morningdailyeco'\n",
        "\n",
        "\n",
        "input_path = BASE_DIR+DATASET+'/vocal_samples'\n",
        "output_path = BASE_DIR+DATASET+'/pitch_analysis'"
      ],
      "metadata": {
        "id": "K5I08m9FXZCS"
      },
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "analyze_folder_and_save_results(input_path, output_path)"
      ],
      "metadata": {
        "id": "ShuwhMCkXcpM"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}