inference.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "id": "initial_id",
   "metadata": {
    "collapsed": true,
    "ExecuteTime": {
     "end_time": "2024-11-25T16:24:30.293039Z",
     "start_time": "2024-11-25T16:24:27.372696Z"
    }
   },
   "source": [
    "import yaml\n",
    "from pathlib import Path\n",
    "\n",
    "import torch\n",
    "from resemblyzer import VoiceEncoder, preprocess_wav\n",
    "import numpy as np\n",
    "from librosa.feature import melspectrogram\n",
    "\n",
    "from common.model_loader import get_model\n",
    "from common.feature_extraction import LogMelFeatureExtractor\n",
    "\n",
    "\n",
    "# path to folder with checkpoint.pth\n",
    "CHECKPOINT_PATH = Path(\"../models/pretrained/100h/PVAD1_SC_DenoisingAPC_finetune_100h_noisy/PVAD1_SC_DenoisingAPC_finetune_100h_noisy_99460/best.pt\").resolve()\n",
    "\n",
    "# Features extractor\n",
    "feature_config = {'sample_rate': 16000,\n",
    "                  'n_mels': 40,\n",
    "                  'n_fft': 400,\n",
    "                  'window_length': 400,\n",
    "                  'hop_length': 160}\n",
    "feature_extractor = LogMelFeatureExtractor(**feature_config)\n",
    "\n",
    "# test device\n",
    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
    "\n",
    "# Get model config\n",
    "yaml.add_multi_constructor('tag:yaml.org,2002:python/object/apply:torch.device',\n",
    "                            lambda loader, suffix, node: None)\n",
    "with open(str(CHECKPOINT_PATH.parent / \"settings.txt\"), \"r\", encoding=\"utf8\") as file:\n",
    "    config = yaml.load(file, Loader=yaml.FullLoader)\n",
    "model_config = config[\"hparams\"][\"model\"][\"encoder\"]\n",
    "\n",
    "# Load model architecture and pretrained weights\n",
    "def load_pretrained_model(model: torch.nn.Module, checkpoint_path: str = \"\", map_location: str = \"cpu\"):\n",
    "    checkpoint = torch.load(checkpoint_path, map_location=map_location)\n",
    "    model_state_dict = model.state_dict()\n",
    "    checkpoint_model_state_dict = checkpoint[\"model_state_dict\"]\n",
    "    model_state_dict.update(checkpoint_model_state_dict)\n",
    "    model.load_state_dict(model_state_dict)\n",
    "    print(f\"Loaded checkpoint {checkpoint_path}.\")\n",
    "    return model\n",
    "model = get_model(model_config)\n",
    "model = load_pretrained_model(model, checkpoint_path=str(CHECKPOINT_PATH), map_location=\"cpu\")\n",
    "model = model.to(device)"
   ],
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded checkpoint /Users/JG96XG/PycharmProjects/SelfSupervisedPVAD/models/pretrained/100h/PVAD1_SC_DenoisingAPC_finetune_100h_noisy/PVAD1_SC_DenoisingAPC_finetune_100h_noisy_99460/best.pt.\n"
     ]
    }
   ],
   "execution_count": 1
  },
  {
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-11-25T16:42:42.573170Z",
     "start_time": "2024-11-25T16:42:42.551484Z"
    }
   },
   "cell_type": "code",
   "source": [
    "speaker_encoder = VoiceEncoder(device=\"cpu\") # For speaker similarity\n",
    "\n",
    "def cos_sim(a, b):\n",
    "    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))\n",
    "\n",
    "def compute_similarity(waveform, speaker_embedding, speaker_encoder):\n",
    "    rate = 2.5\n",
    "    min_coverage = 0.5\n",
    "    waveform_copy = waveform[0].numpy().copy()\n",
    "    wav_slices, mel_slices = speaker_encoder.compute_partial_slices(waveform_copy.size,\n",
    "                                                                    rate=rate,\n",
    "                                                                    min_coverage=min_coverage)\n",
    "    max_wave_length = wav_slices[-1].stop\n",
    "    if max_wave_length >= waveform_copy.size:\n",
    "        waveform_copy = np.pad(waveform_copy, (0, max_wave_length - waveform_copy.size), \"constant\")\n",
    "    features = melspectrogram(y=waveform_copy, sr=16000, n_fft=400, hop_length=160, n_mels=40).astype('float32').T\n",
    "    features_sliced = np.array([features[s] for s in mel_slices])\n",
    "    with torch.no_grad():\n",
    "        embedding_slices = speaker_encoder(torch.from_numpy(features_sliced)).numpy()\n",
    "\n",
    "    scores_slices = np.array([cos_sim(speaker_embedding, partial_embedding) for partial_embedding in embedding_slices])\n",
    "\n",
    "    # Scores, linearly interpolated, starting from 0.5 every time\n",
    "    frame_step = int(np.round((16000 / rate) / 160))\n",
    "    # First 160 frames are score one (160 frames window)\n",
    "    scores_linear_interpolated = np.kron(scores_slices[0], np.ones(160, dtype='float32'))\n",
    "    # The rest are linearly interpolated\n",
    "    for i, s in enumerate(scores_slices[1:]):\n",
    "        scores_linear_interpolated = np.append(scores_linear_interpolated,\n",
    "                                               np.linspace(scores_slices[i], s, frame_step, endpoint=False))\n",
    "\n",
    "    similarity = torch.from_numpy(scores_linear_interpolated)\n",
    "    return similarity"
   ],
   "id": "337cf85ef436e326",
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded the voice encoder model on cpu in 0.01 seconds.\n"
     ]
    }
   ],
   "execution_count": 49
  },
  {
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-11-25T16:46:31.895697Z",
     "start_time": "2024-11-25T16:46:31.568852Z"
    }
   },
   "cell_type": "code",
   "source": [
    "# Example input\n",
    "input_tensor = torch.randn(size=(1, 160000))\n",
    "reference_tensor = torch.randn(size=(1, 160000))\n",
    "\n",
    "# Compute speaker embedding of reference speech\n",
    "waveform_preprocessed = preprocess_wav(reference_tensor[0].numpy())\n",
    "speaker_embedding = torch.from_numpy(speaker_encoder.embed_speaker([waveform_preprocessed])).unsqueeze(0)"
   ],
   "id": "21c96045f2868b5f",
   "outputs": [],
   "execution_count": 82
  },
  {
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-11-25T16:49:31.466736Z",
     "start_time": "2024-11-25T16:49:31.133561Z"
    }
   },
   "cell_type": "code",
   "source": [
    "# Compute features\n",
    "features = feature_extractor(input_tensor)\n",
    "lengths = torch.tensor([features.size(-1)])\n",
    "\n",
    "# Compute similarity vector\n",
    "similarity = compute_similarity(input_tensor, speaker_embedding, speaker_encoder)[:features.size(-1)]\n",
    "\n",
    "features = features.transpose(1, 2)\n",
    "similarity = similarity.unsqueeze(0)"
   ],
   "id": "c1da32b34a4216c1",
   "outputs": [],
   "execution_count": 100
  },
  {
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-11-25T16:49:31.474142Z",
     "start_time": "2024-11-25T16:49:31.472101Z"
    }
   },
   "cell_type": "code",
   "source": [
    "print(similarity.size())\n",
    "print(features.size())"
   ],
   "id": "b34cd04b6f49654e",
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([1, 1001])\n",
      "torch.Size([1, 1001, 40])\n"
     ]
    }
   ],
   "execution_count": 101
  },
  {
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-11-25T16:51:43.231602Z",
     "start_time": "2024-11-25T16:51:43.199992Z"
    }
   },
   "cell_type": "code",
   "source": [
    "# Put input through model\n",
    "with torch.no_grad():\n",
    "    outputs = model(x=features, x_lens=lengths, similarity_scores=similarity)"
   ],
   "id": "ed717a4f0993fb25",
   "outputs": [],
   "execution_count": 106
  },
  {
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-11-25T16:52:58.766887Z",
     "start_time": "2024-11-25T16:52:58.762440Z"
    }
   },
   "cell_type": "code",
   "source": "vad_scores = torch.nn.functional.softmax(outputs[0], dim=-1)",
   "id": "24d7bf538b1828ff",
   "outputs": [],
   "execution_count": 112
  },
  {
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-11-25T16:53:09.609796Z",
     "start_time": "2024-11-25T16:53:09.517793Z"
    }
   },
   "cell_type": "code",
   "source": [
    "from matplotlib import pyplot as plt\n",
    "\n",
    "plt.plot(vad_scores[0, :, 0], label=\"no-speech\")\n",
    "plt.plot(vad_scores[0, :, 1], label=\"target-speech\")\n",
    "plt.plot(vad_scores[0, :, 2], label=\"non-target-speech\")\n",
    "plt.legend()\n",
    "plt.show()"
   ],
   "id": "616c2d9fc9d0c862",
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ],
      "image/png": "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"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "execution_count": 113
  },
  {
   "metadata": {},
   "cell_type": "code",
   "outputs": [],
   "execution_count": null,
   "source": "",
   "id": "fff50a472cebe29c"
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}