Added jupyter notebook of OrpheusTTS used to tokenize Thorsten-Voice dataset.

OrpheusTTS_Tokenize_NB_used_by_Thorsten_Voice.ipynb

@@ -0,0 +1,309 @@
+{
+  "nbformat": 4,
+  "nbformat_minor": 0,
+  "metadata": {
+    "colab": {
+      "provenance": [],
+      "machine_shape": "hm",
+      "gpuType": "T4"
+    },
+    "kernelspec": {
+      "name": "python3",
+      "display_name": "Python 3"
+    },
+    "language_info": {
+      "name": "python"
+    },
+    "accelerator": "GPU"
+  },
+  "cells": [
+    {
+      "cell_type": "code",
+      "source": [
+        "my_original_dataset_name = \"Thorsten-Voice/TV-24kHz-2025.12-Neutral-FT-Mini\"\n",
+        "name_to_push_dataset_to = \"Thorsten-Voice/TV-24kHz-2025.12-Neutral-FT-Mini-tokenised\"\n",
+        "!huggingface-cli login --token=SECRET"
+      ],
+      "metadata": {
+        "id": "5uX_IoEpnnL2"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "!pip install torchcodec\n",
+        "!pip install datasets==3.5.1 # Using datasets >= 4.0 had some issues with this codebase"
+      ],
+      "metadata": {
+        "id": "4Y6bf-Kgxpz2"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "Y1BlCraIs9bh"
+      },
+      "outputs": [],
+      "source": [
+        "#@title Installation & Setup\n",
+        "#%%capture\n",
+        "import locale\n",
+        "locale.getpreferredencoding = lambda: \"UTF-8\"\n",
+        "!pip install datasets==3.5.1\n",
+        "!pip install snac\n",
+        "import torch\n",
+        "import torchcodec\n",
+        "from snac import SNAC\n",
+        "from datasets import load_dataset\n",
+        "from huggingface_hub import snapshot_download\n",
+        "from datasets import load_dataset\n",
+        "\n",
+        "dsn = my_original_dataset_name\n",
+        "\n",
+        "snapshot_download(\n",
+        "    repo_id=dsn,\n",
+        "    repo_type=\"dataset\",\n",
+        "    revision=\"main\",\n",
+        "    max_workers=64,\n",
+        ")\n",
+        "\n",
+        "\n",
+        "ds = load_dataset(dsn, split=\"train\")\n",
+        "ds_sample_rate = ds[0][\"audio\"][\"sampling_rate\"]\n",
+        "\n",
+        "print(ds_sample_rate)\n",
+        "\n",
+        "model = SNAC.from_pretrained(\"hubertsiuzdak/snac_24khz\")\n",
+        "model = model.to(\"cuda\")"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "# Just for testing purpose, to check if sample rate is correct on Thorsten-Voice dataset\n",
+        "ds[0][\"audio\"][\"sampling_rate\"]"
+      ],
+      "metadata": {
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        },
+        "id": "l_ycyGL_7X70",
+        "outputId": "1002b1a6-be2e-42d2-cde9-19eac01c0beb"
+      },
+      "execution_count": null,
+      "outputs": [
+        {
+          "output_type": "execute_result",
+          "data": {
+            "text/plain": [
+              "24000"
+            ]
+          },
+          "metadata": {},
+          "execution_count": 5
+        }
+      ]
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "#@title Tokenisation Function\n",
+        "import torchaudio.transforms as T\n",
+        "def tokenise_audio(waveform):\n",
+        "  waveform = torch.from_numpy(waveform).unsqueeze(0)\n",
+        "  waveform = waveform.to(dtype=torch.float32)\n",
+        "  resample_transform = T.Resample(orig_freq=ds_sample_rate, new_freq=24000)\n",
+        "  waveform = resample_transform(waveform)\n",
+        "\n",
+        "  waveform = waveform.unsqueeze(0).to(\"cuda\")\n",
+        "\n",
+        "  #generate the codes from snac\n",
+        "  with torch.inference_mode():\n",
+        "    codes = model.encode(waveform)\n",
+        "\n",
+        "  all_codes = []\n",
+        "  for i in range(codes[0].shape[1]):\n",
+        "    all_codes.append(codes[0][0][i].item()+128266)\n",
+        "    all_codes.append(codes[1][0][2*i].item()+128266+4096)\n",
+        "    all_codes.append(codes[2][0][4*i].item()+128266+(2*4096))\n",
+        "    all_codes.append(codes[2][0][(4*i)+1].item()+128266+(3*4096))\n",
+        "    all_codes.append(codes[1][0][(2*i)+1].item()+128266+(4*4096))\n",
+        "    all_codes.append(codes[2][0][(4*i)+2].item()+128266+(5*4096))\n",
+        "    all_codes.append(codes[2][0][(4*i)+3].item()+128266+(6*4096))\n",
+        "\n",
+        "\n",
+        "  return all_codes\n",
+        "\n",
+        "\n"
+      ],
+      "metadata": {
+        "id": "kbZENwXltYSC"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "#@title Map Tokenize\n",
+        "import random\n",
+        "def add_codes(example):\n",
+        "    # Always initialize codes_list to None\n",
+        "    codes_list = None\n",
+        "\n",
+        "    print(example.get(\"audio\"))\n",
+        "\n",
+        "\n",
+        "    try:\n",
+        "        answer_audio = example.get(\"audio\")\n",
+        "        # If there's a valid audio array, tokenise it\n",
+        "        if answer_audio and \"array\" in answer_audio:\n",
+        "            audio_array = answer_audio[\"array\"]\n",
+        "            codes_list = tokenise_audio(audio_array)\n",
+        "    except Exception as e:\n",
+        "        print(f\"Skipping row due to error: {e}\")\n",
+        "        # Keep codes_list as None if we fail\n",
+        "    example[\"codes_list\"] = codes_list\n",
+        "\n",
+        "    return example\n",
+        "\n",
+        "ds = ds.map(add_codes, remove_columns=[\"audio\"])\n"
+      ],
+      "metadata": {
+        "id": "Yv9OPDpRwWOy"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "#@title Load Tokenizer\n",
+        "tokeniser_length = 128256\n",
+        "start_of_text = 128000\n",
+        "end_of_text = 128009\n",
+        "\n",
+        "start_of_speech = tokeniser_length + 1\n",
+        "end_of_speech = tokeniser_length + 2\n",
+        "\n",
+        "start_of_human = tokeniser_length + 3\n",
+        "end_of_human = tokeniser_length + 4\n",
+        "\n",
+        "start_of_ai = tokeniser_length + 5\n",
+        "end_of_ai =  tokeniser_length + 6\n",
+        "pad_token = tokeniser_length + 7\n",
+        "\n",
+        "audio_tokens_start = tokeniser_length + 10\n",
+        "\n",
+        "tokenizer_name = \"canopylabs/orpheus-3b-0.1-pretrained\"\n",
+        "\n",
+        "from transformers import AutoTokenizer\n",
+        "import os\n",
+        "tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)\n",
+        "num_proc = os.cpu_count()\n",
+        "\n",
+        "ds = ds.filter(lambda x: x[\"codes_list\"] is not None)\n",
+        "ds = ds.filter(lambda x: len(x[\"codes_list\"]) > 0)"
+      ],
+      "metadata": {
+        "id": "2G9uppg0H3-X"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "#@title Create Input Ids\n",
+        "def remove_duplicate_frames(example):\n",
+        "    vals = example[\"codes_list\"]\n",
+        "    if len(vals) % 7 != 0:\n",
+        "        raise ValueError(\"Input list length must be divisible by 7\")\n",
+        "\n",
+        "    result = vals[:7]\n",
+        "\n",
+        "    removed_frames = 0\n",
+        "\n",
+        "    for i in range(7, len(vals), 7):\n",
+        "        current_first = vals[i]\n",
+        "        previous_first = result[-7]\n",
+        "\n",
+        "        if current_first != previous_first:\n",
+        "            result.extend(vals[i:i+7])\n",
+        "        else:\n",
+        "            removed_frames += 1\n",
+        "\n",
+        "    example[\"codes_list\"] = result\n",
+        "\n",
+        "    return example\n",
+        "\n",
+        "ds = ds.map(remove_duplicate_frames, num_proc=num_proc)\n",
+        "\n",
+        "tok_info = '''*** HERE you can modify the text prompt\n",
+        "i.e. if you wanted a multispeaker model like canopylabs/orpheus-3b-0.1-ft, you can pass:\n",
+        "f\"{example[\"source\"]}:  {example[\"text\"]}\", as is passed.\n",
+        "'''\n",
+        "print(tok_info)\n",
+        "\n",
+        "def create_input_ids(example):\n",
+        "    text_ids = tokenizer.encode(example[\"text\"],  add_special_tokens=True)\n",
+        "    text_ids.append(end_of_text)\n",
+        "    example[\"text_tokens\"] = text_ids\n",
+        "    input_ids = (\n",
+        "        [start_of_human]\n",
+        "        + example[\"text_tokens\"]\n",
+        "        + [end_of_human]\n",
+        "        + [start_of_ai]\n",
+        "        + [start_of_speech]\n",
+        "        + example[\"codes_list\"]\n",
+        "        + [end_of_speech]\n",
+        "        + [end_of_ai]\n",
+        "    )\n",
+        "    example[\"input_ids\"] = input_ids\n",
+        "    example[\"labels\"] = input_ids\n",
+        "    example[\"attention_mask\"] = [1] * len(input_ids)\n",
+        "\n",
+        "    return example\n",
+        "\n",
+        "ds = ds.map(create_input_ids, num_proc=num_proc, remove_columns=[\"text\", \"codes_list\"])\n"
+      ],
+      "metadata": {
+        "id": "hWGtOc5QIPcn"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "#@title Remove unnecessary columns\n",
+        "columns_to_keep = [\"input_ids\", \"labels\", \"attention_mask\"]\n",
+        "columns_to_remove = [col for col in ds.column_names if col not in columns_to_keep]\n",
+        "\n",
+        "ds = ds.remove_columns(columns_to_remove)"
+      ],
+      "metadata": {
+        "id": "ee3zbdCUIWV6"
+      },
+      "execution_count": null,
+      "outputs": []
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "ame_to_push_dataset_to = \"Thorsten-Voice/TV-24kHz-2025.12-Neutral-FT-Mini-tokenised\"\n",
+        "ds.push_to_hub(name_to_push_dataset_to)"
+      ],
+      "metadata": {
+        "id": "Ov_2ItW6nldr"
+      },
+      "execution_count": null,
+      "outputs": []
+    }
+  ]
+}