{ "cells": [ { "cell_type": "markdown", "metadata": { "id": "view-in-github", "colab_type": "text" }, "source": [ "\"Open" ] }, { "cell_type": "markdown", "source": [ "# A two-stage U-Net for high-fidelity denoising of historical recordings" ], "metadata": { "id": "jbe_aWYkjWRH" }, "id": "jbe_aWYkjWRH" }, { "cell_type": "markdown", "source": [ "This notebook is a demo of the historical music denoising method proposed in:\n", "\n", "> E. Moliner and V. Välimäki,, \"A two-stage U-Net for high-fidelity denosing of historical recordings\", submitted to IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Singapore, May, 2022\n", "\n", "

\n", "\"Schema

\n", "\n", "Listen to our [audio samples](http://research.spa.aalto.fi/publications/papers/icassp22-denoising/)\n", "\n", "You can freely use it to denoise your own historical recordings.\n", "\n", "### Instructions for running:\n", "\n", "* Make sure to use a GPU runtime, click: __Runtime >> Change Runtime Type >> GPU__\n", "* Press ▶️ on the left of each of the cells\n", "* View the code: Double-click any of the cells\n", "* Hide the code: Double click the right side of the cell\n", "* For some reason, this notebook does not work in Firefox, so please use another browser.\n" ], "metadata": { "id": "8UON6ncSApA9" }, "id": "8UON6ncSApA9" }, { "cell_type": "code", "source": [ "#@title #Install and Import\n", "\n", "#@markdown Execute this cell to install the required data and dependencies. This step might take some time.\n", "\n", "#download the files\n", "! git clone https://github.com/eloimoliner/denoising-historical-recordings.git\n", "! wget https://github.com/eloimoliner/denoising-historical-recordings/releases/download/v0.0/checkpoint.zip\n", "! unzip checkpoint.zip -d denoising-historical-recordings/experiments/trained_model/\n", "\n", "%cd denoising-historical-recordings\n", "\n", "#install dependencies\n", "! pip install hydra-core==0.11.3\n", "\n", "#All the code goes here\n", "import unet\n", "import tensorflow as tf\n", "import soundfile as sf\n", "import numpy as np\n", "from tqdm import tqdm\n", "import scipy.signal\n", "import hydra\n", "import os\n", "#workaround to load hydra conf file\n", "import yaml\n", "from pathlib import Path\n", "args = yaml.safe_load(Path('conf/conf.yaml').read_text())\n", "class dotdict(dict):\n", " \"\"\"dot.notation access to dictionary attributes\"\"\"\n", " __getattr__ = dict.get\n", " __setattr__ = dict.__setitem__\n", " __delattr__ = dict.__delitem__\n", "args=dotdict(args)\n", "unet_args=dotdict(args.unet)\n", "\n", "path_experiment=str(args.path_experiment)\n", "\n", "unet_model = unet.build_model_denoise(unet_args=unet_args)\n", "\n", "ckpt=os.path.join(\"/content/denoising-historical-recordings\",path_experiment, 'checkpoint')\n", "unet_model.load_weights(ckpt)\n", "\n", "def do_stft(noisy):\n", " \n", " window_fn = tf.signal.hamming_window\n", "\n", " win_size=args.stft[\"win_size\"]\n", " hop_size=args.stft[\"hop_size\"]\n", "\n", " \n", " stft_signal_noisy=tf.signal.stft(noisy,frame_length=win_size, window_fn=window_fn, frame_step=hop_size, pad_end=True)\n", " stft_noisy_stacked=tf.stack( values=[tf.math.real(stft_signal_noisy), tf.math.imag(stft_signal_noisy)], axis=-1)\n", "\n", " return stft_noisy_stacked\n", "\n", "def do_istft(data):\n", " \n", " window_fn = tf.signal.hamming_window\n", "\n", " win_size=args.stft[\"win_size\"]\n", " hop_size=args.stft[\"hop_size\"]\n", "\n", " inv_window_fn=tf.signal.inverse_stft_window_fn(hop_size, forward_window_fn=window_fn)\n", "\n", " pred_cpx=data[...,0] + 1j * data[...,1]\n", " pred_time=tf.signal.inverse_stft(pred_cpx, win_size, hop_size, window_fn=inv_window_fn)\n", " return pred_time\n", "\n", "def denoise_audio(audio):\n", "\n", " data, samplerate = sf.read(audio)\n", " print(data.dtype)\n", " #Stereo to mono\n", " if len(data.shape)>1:\n", " data=np.mean(data,axis=1)\n", " \n", " if samplerate!=44100: \n", " print(\"Resampling\")\n", " \n", " data=scipy.signal.resample(data, int((44100 / samplerate )*len(data))+1) \n", " \n", " \n", " \n", " segment_size=44100*5 #20s segments\n", "\n", " length_data=len(data)\n", " overlapsize=2048 #samples (46 ms)\n", " window=np.hanning(2*overlapsize)\n", " window_right=window[overlapsize::]\n", " window_left=window[0:overlapsize]\n", " audio_finished=False\n", " pointer=0\n", " denoised_data=np.zeros(shape=(len(data),))\n", " residual_noise=np.zeros(shape=(len(data),))\n", " numchunks=int(np.ceil(length_data/segment_size))\n", " \n", " for i in tqdm(range(numchunks)):\n", " if pointer+segment_size\n", " \n", " Upload widget is only available when the cell has been executed in the\n", " current browser session. Please rerun this cell to enable.\n", " \n", " " ], "text/plain": [ "" ] }, "metadata": {} }, { "output_type": "stream", "name": "stdout", "text": [ "Saving Carmen-Habanera_(Love_is_Like_a_Woo_-_Marguerite_D'Alvarez_noisy_input.wav to Carmen-Habanera_(Love_is_Like_a_Woo_-_Marguerite_D'Alvarez_noisy_input.wav\n" ] } ] }, { "cell_type": "code", "source": [ "#@title #Denoise\n", "\n", "#@markdown Execute this cell to denoise the uploaded file\n", "for fn in uploaded.keys():\n", " print('Denoising uploaded file \"{name}\"'.format(\n", " name=fn))\n", " denoise_data=denoise_audio(fn)\n", " basename=os.path.splitext(fn)[0]\n", " wav_output_name=basename+\"_denoised\"+\".wav\"\n", " sf.write(wav_output_name, denoise_data, 44100)" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "cellView": "form", "id": "0po6zpvrylc2", "outputId": "173f5355-2939-41fe-c702-591aa752fc7e" }, "id": "0po6zpvrylc2", "execution_count": null, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Denoising uploaded file \"Carmen-Habanera_(Love_is_Like_a_Woo_-_Marguerite_D'Alvarez_noisy_input.wav\"\n", "float64\n" ] }, { "output_type": "stream", "name": "stderr", "text": [ "100%|██████████| 41/41 [00:30<00:00, 1.34it/s]\n" ] } ] }, { "cell_type": "code", "source": [ "#@title #Download\n", "\n", "#@markdown Execute this cell to download the denoised recording\n", "files.download(wav_output_name)" ], "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 17 }, "cellView": "form", "id": "3tEshWBezYvf", "outputId": "54588c26-0b3c-42bf-aca2-8316ab54603f" }, "id": "3tEshWBezYvf", "execution_count": null, "outputs": [ { "output_type": "display_data", "data": { "application/javascript": [ "\n", " async function download(id, filename, size) {\n", " if (!google.colab.kernel.accessAllowed) {\n", " return;\n", " }\n", " const div = document.createElement('div');\n", " const label = document.createElement('label');\n", " label.textContent = `Downloading \"${filename}\": `;\n", " div.appendChild(label);\n", " const progress = document.createElement('progress');\n", " progress.max = size;\n", " div.appendChild(progress);\n", " document.body.appendChild(div);\n", "\n", " const buffers = [];\n", " let downloaded = 0;\n", "\n", " const channel = await google.colab.kernel.comms.open(id);\n", " // Send a message to notify the kernel that we're ready.\n", " channel.send({})\n", "\n", " for await (const message of channel.messages) {\n", " // Send a message to notify the kernel that we're ready.\n", " channel.send({})\n", " if (message.buffers) {\n", " for (const buffer of message.buffers) {\n", " buffers.push(buffer);\n", " downloaded += buffer.byteLength;\n", " progress.value = downloaded;\n", " }\n", " }\n", " }\n", " const blob = new Blob(buffers, {type: 'application/binary'});\n", " const a = document.createElement('a');\n", " a.href = window.URL.createObjectURL(blob);\n", " a.download = filename;\n", " div.appendChild(a);\n", " a.click();\n", " div.remove();\n", " }\n", " " ], "text/plain": [ "" ] }, "metadata": {} }, { "output_type": "display_data", "data": { "application/javascript": [ "download(\"download_3e3ca242-937f-408e-a16c-4aa5bb2b9e50\", \"Carmen-Habanera_(Love_is_Like_a_Woo_-_Marguerite_D'Alvarez_noisy_input_denoised.wav\", 17961334)" ], "text/plain": [ "" ] }, "metadata": {} } ] }, { "cell_type": "code", "source": [ "" ], "metadata": { "id": "v_FuSJ4J-WO-" }, "id": "v_FuSJ4J-WO-", "execution_count": null, "outputs": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.8.5" }, "colab": { "name": "demo.ipynb", "provenance": [], "include_colab_link": true }, "accelerator": "GPU" }, "nbformat": 4, "nbformat_minor": 5 }