{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import pandas as pd\n", "\n", "filepath = \"/home/tony/Data/Preference/30b_v3/interesting_clips_v4_t_4_20240923_full_with_sem_distance_and_similarity.pkl\"\n", "train_df = pd.read_pickle(filepath)\n", "train_df = train_df.sort_values(by=[\"request_id\", \"preference\"])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "train_df.tail(n=2)[[\"s3_id\", \"prompt_text\", \"similarity\", \"preference\"]]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import os\n", "from suno_utils.audio import Audio\n", "from suno_utils.tasks.mert_25 import preload_models, encode\n", "_ = preload_models( \n", " checkpoint_filepath=\"s3://suno-data/georg/models/semantic/mert_25.pt\",\n", " centroids_filepath=\"s3://suno-data/georg/models/semantic/mert_25_2x4k.npy\",\n", " device=\"cuda\",\n", ")\n", "\n", "\n", "def measure_similarity(\n", " input_s3_id: str, \n", " cover_a_s3_id: str, \n", " cover_b_s3_id: str, \n", " s3_base_path: str = \"s3://suno-data-uploads/studio/uploads/\", \n", " feature: str = \"semantic\"\n", " ):\n", " input_audio = Audio.from_s3(os.path.join(s3_base_path, input_s3_id + \".mp3\"))\n", " cover_a_audio = Audio.from_s3(os.path.join(s3_base_path, cover_a_s3_id + \".mp3\"))\n", " cover_b_audio = Audio.from_s3(os.path.join(s3_base_path, cover_b_s3_id + \".mp3\"))\n", "\n", " if feature == \"semantic\":\n", " input_sem = encode(input_audio, do_clustering=False)\n", " cover_a_sem = encode(cover_a_audio, do_clustering=False)\n", " cover_b_sem = encode(cover_b_audio, do_clustering=False)\n", "\n", " input_mean = input_sem.mean(axis=0)\n", " cover_a_mean = cover_a_sem.mean(axis=0)\n", " cover_b_mean = cover_b_sem.mean(axis=0)\n", " elif feature == \"crema\":\n", " input_crema = analyze(y=input_audio.float_array, sr=input_audio.sample_rate)\n", " cover_a_crema = analyze(y=cover_a_audio.float_array, sr=cover_a_audio.sample_rate)\n", " cover_b_crema = analyze(y=cover_b_audio.float_array, sr=cover_b_audio.sample_rate)\n", " print(input_crema.shape)\n", "\n", " input_mean = input_crema.mean(axis=0)\n", " cover_a_mean = cover_a_crema.mean(axis=0)\n", " cover_b_mean = cover_b_crema.mean(axis=0)\n", " else:\n", " raise ValueError(f\"Feature {feature} not supported\")\n", "\n", " # mse distance between input and cover_a\n", " input_to_cover_a_distance = ((input_mean - cover_a_mean) ** 2).sum()\n", " input_to_cover_b_distance = ((input_mean - cover_b_mean) ** 2).sum()\n", "\n", " return input_to_cover_a_distance, input_to_cover_b_distance\n", "\n", "\n", "def analyze_row(index: int):\n", " # get the cover id \n", " row1_metadata = dict(train_df.iloc[index][\"metadata\"])\n", " input_s3_id = row1_metadata[\"cover_clip_id\"]\n", " cover_s3_id = train_df.iloc[index][\"s3_id\"]\n", " s3_base_path: str = \"s3://suno-data-uploads/studio/uploads/\"\n", "\n", " input_audio = Audio.from_s3(os.path.join(s3_base_path, input_s3_id + \".mp3\"))\n", " cover_audio = Audio.from_s3(os.path.join(s3_base_path, cover_s3_id + \".mp3\"))\n", "\n", " input_sem = encode(input_audio, do_clustering=False)\n", " cover_sem = encode(cover_audio, do_clustering=False)\n", "\n", " input_mean = input_sem.mean(axis=0)\n", " cover_mean = cover_sem.mean(axis=0)\n", "\n", " # mse distance between input and cover\n", " input_to_cover_distance = ((input_mean - cover_mean) ** 2).sum()\n", "\n", " return input_to_cover_distance" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# print all of the key names \n", "print(train_df.iloc[0].keys())\n", "\n", "# put the metadata object into a dictionary\n", "cover_id = train_df.iloc[0][\"s3_id\"]\n", "metadata = train_df.iloc[0][\"metadata\"]\n", "metadata_dict = dict(metadata)\n", "print(cover_id)\n", "print(metadata_dict[\"cover_clip_id\"])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# iterate over all rows and measure similarity between cover a and cover b\n", "# one example is comprised of two adjacent rows in the dataframe\n", "# so iterate over the dataframe in steps of 2 and get two rows at a time\n", "for index in range(0, len(train_df), 2):\n", " print(f\"Processing index {index}\")\n", " row1 = train_df.iloc[index]\n", " row2 = train_df.iloc[index + 1]\n", "\n", " a_tony_sim = row1[\"similarity\"]\n", " b_tony_sim = row2[\"similarity\"]\n", " a_preference = row1[\"preference\"]\n", " b_preference = row2[\"preference\"]\n", "\n", " # get the cover id \n", " row1_metadata = dict(row1[\"metadata\"])\n", " input_s3_id = row1_metadata[\"cover_clip_id\"]\n", "\n", " cover_a_s3_id = row1[\"s3_id\"]\n", " cover_b_s3_id = row2[\"s3_id\"]\n", "\n", " input_to_cover_a_distance, input_to_cover_b_distance = measure_similarity(input_s3_id, cover_a_s3_id, cover_b_s3_id)\n", " print(f\"a dist: {input_to_cover_a_distance}, b dist: {input_to_cover_b_distance}\")\n", " print(f\"a tony sim: {a_tony_sim}, b tony sim: {b_tony_sim}\")\n", " print(f\"a preference: {a_preference}, b preference: {b_preference}\")\n", " print(\"\")\n", "\n", " # add similarity back to the dataframe\n", " train_df.at[index, \"distance\"] = input_to_cover_a_distance\n", " train_df.at[index + 1, \"distance\"] = input_to_cover_b_distance\n", "\n", " if index > 10:\n", " break" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "suno_env", "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": "3.10.9" } }, "nbformat": 4, "nbformat_minor": 2 }