{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "#### Speaker verification\n", "\n", "```python\n", "import torch\n", "from pyannote.audio.pipelines.speaker_verification import PretrainedSpeakerEmbedding\n", "model = PretrainedSpeakerEmbedding(\n", " \"speechbrain/spkrec-ecapa-voxceleb\",\n", " device=torch.device(\"cuda\"))\n", "\n", "from pyannote.audio import Audio\n", "from pyannote.core import Segment\n", "audio = Audio(sample_rate=16000, mono=True)\n", "\n", "# extract embedding for a speaker speaking between t=3s and t=6s\n", "speaker1 = Segment(3., 6.)\n", "waveform1, sample_rate = audio.crop(\"audio.wav\", speaker1)\n", "embedding1 = model(waveform1[None])\n", "\n", "# extract embedding for a speaker speaking between t=7s and t=12s\n", "speaker2 = Segment(7., 12.)\n", "waveform2, sample_rate = audio.crop(\"audio.wav\", speaker2)\n", "embedding2 = model(waveform2[None])\n", "\n", "# compare embeddings using \"cosine\" distance\n", "from scipy.spatial.distance import cdist\n", "distance = cdist(embedding1, embedding2, metric=\"cosine\")\n", "```\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [] } ], "metadata": { "interpreter": { "hash": "41379f2c2a4eb17f5ac9a1f5014f4b793a0ead0b6469d8877f81a91eb030f53e" }, "kernelspec": { "display_name": "Python 3.8.2 64-bit ('pyannote': conda)", "language": "python", "name": "python3" }, "language_info": { "name": "python", "version": "3.8.2" } }, "nbformat": 4, "nbformat_minor": 2 }