diff --git a/notebooks/piper_inference_(ckpt).ipynb b/notebooks/piper_inference_(ckpt).ipynb index e106c0a..24bbad7 100644 --- a/notebooks/piper_inference_(ckpt).ipynb +++ b/notebooks/piper_inference_(ckpt).ipynb @@ -5,7 +5,7 @@ "colab": { "provenance": [], "gpuType": "T4", - "authorship_tag": "ABX9TyMcevzeVyewWF1ZHKzBu3CB", + "authorship_tag": "ABX9TyNju0yzRK8wgAS+WgyeTEAl", "include_colab_link": true }, "kernelspec": { @@ -88,12 +88,13 @@ " playaudio(\"installing\")\n", "!git clone -q https://github.com/rmcpantoja/piper\n", "%cd /content/piper/src/python\n", - "!pip install -q -r requirements.txt\n", + "#!pip install -q -r requirements.txt\n", + "!pip install -q cython>=0.29.0 piper-phonemize==1.1.0 librosa>=0.9.2 numpy>=1.19.0 onnxruntime>=1.11.0 pytorch-lightning==1.7.0 torch==1.11.0\n", "!pip install -q torchtext==0.12.0 torchvision==0.12.0\n", + "#!pip install -q torchtext==0.14.1 torchvision==0.14.1\n", "# fixing recent compativility isswes:\n", "!pip install -q torchaudio==0.11.0 torchmetrics==0.11.4\n", "!bash build_monotonic_align.sh\n", - "!apt-get install -q espeak-ng\n", "import os\n", "if not os.path.exists(\"/content/piper/src/python/lng\"):\n", " !cp -r \"/content/piper/notebooks/lng\" /content/piper/src/python/lng\n", @@ -190,6 +191,8 @@ "import logging\n", "import sys\n", "from pathlib import Path\n", + "from enum import Enum\n", + "from typing import Iterable, List, Optional, Union\n", "import torch\n", "from piper_train.vits.lightning import VitsModel\n", "from piper_train.vits.utils import audio_float_to_int16\n", @@ -198,8 +201,7 @@ "import glob\n", "import ipywidgets as widgets\n", "from IPython.display import display, Audio, Markdown, clear_output\n", - "from espeak_phonemizer import Phonemizer\n", - "from piper_train import phonemize\n", + "from piper_phonemize import phonemize_codepoints, phonemize_espeak, tashkeel_run\n", "\n", "_LOGGER = logging.getLogger(\"piper_train.infer_onnx\")\n", "\n", @@ -382,35 +384,69 @@ " config = json.load(file)\n", " return config\n", "\n", + "PAD = \"_\" # padding (0)\n", + "BOS = \"^\" # beginning of sentence\n", + "EOS = \"$\" # end of sentence\n", + "\n", + "class PhonemeType(str, Enum):\n", + " ESPEAK = \"espeak\"\n", + " TEXT = \"text\"\n", + "\n", + "def phonemize(config, text: str) -> List[List[str]]:\n", + " \"\"\"Text to phonemes grouped by sentence.\"\"\"\n", + " if config[\"phoneme_type\"] == PhonemeType.ESPEAK:\n", + " if config[\"espeak\"][\"voice\"] == \"ar\":\n", + " # Arabic diacritization\n", + " # https://github.com/mush42/libtashkeel/\n", + " text = tashkeel_run(text)\n", + " return phonemize_espeak(text, config[\"espeak\"][\"voice\"])\n", + " if config[\"phoneme_type\"] == PhonemeType.TEXT:\n", + " return phonemize_codepoints(text)\n", + " raise ValueError(f\"Unexpected phoneme type: {self.config.phoneme_type}\")\n", + "\n", + "def phonemes_to_ids(config, phonemes: List[str]) -> List[int]:\n", + " \"\"\"Phonemes to ids.\"\"\"\n", + " id_map = config[\"phoneme_id_map\"]\n", + " ids: List[int] = list(id_map[BOS])\n", + " for phoneme in phonemes:\n", + " if phoneme not in id_map:\n", + " print(\"Missing phoneme from id map: %s\", phoneme)\n", + " continue\n", + " ids.extend(id_map[phoneme])\n", + " ids.extend(id_map[PAD])\n", + " ids.extend(id_map[EOS])\n", + " return ids\n", + "\n", "def inferencing(model, config, sid, line, length_scale = 1, noise_scale = 0.667, noise_scale_w = 0.8, auto_play=True):\n", - " espeak_voice = config[\"espeak\"][\"voice\"]\n", - " phonemizer = Phonemizer(default_voice=espeak_voice)\n", - " phonemes = phonemize.phonemize(line, phonemizer)\n", - " ids = phonemize.phonemes_to_ids(phonemes)\n", - " phoneme_ids = ids\n", - " num_speakers = config[\"num_speakers\"]\n", - " if num_speakers == 1:\n", - " speaker_id = None # for now\n", - " else:\n", - " speaker_id = sid\n", - " text = torch.LongTensor(phoneme_ids).unsqueeze(0)\n", - " text_lengths = torch.LongTensor([len(phoneme_ids)])\n", - " scales = [\n", - " noise_scale,\n", - " length_scale,\n", - " noise_scale_w\n", - " ]\n", - " sid = torch.LongTensor([speaker_id]) if speaker_id is not None else None\n", - " audio = model(\n", - " text,\n", - " text_lengths,\n", - " scales,\n", - " sid=sid\n", - " ).detach().numpy()\n", - " audio = audio_float_to_int16(audio.squeeze())\n", + " audios = []\n", + " text = phonemize(config, line)\n", + " for phonemes in text:\n", + " phoneme_ids = phonemes_to_ids(config, phonemes)\n", + " num_speakers = config[\"num_speakers\"]\n", + " if num_speakers == 1:\n", + " speaker_id = None # for now\n", + " else:\n", + " speaker_id = sid\n", + " text = torch.LongTensor(phoneme_ids).unsqueeze(0)\n", + " text_lengths = torch.LongTensor([len(phoneme_ids)])\n", + " scales = [\n", + " noise_scale,\n", + " length_scale,\n", + " noise_scale_w\n", + " ]\n", + " sid = torch.LongTensor([speaker_id]) if speaker_id is not None else None\n", + " audio = model(\n", + " text,\n", + " text_lengths,\n", + " scales,\n", + " sid=sid\n", + " ).detach().numpy()\n", + " audio = audio_float_to_int16(audio.squeeze())\n", + " audios.append(audio)\n", + " merged_audio = np.concatenate(audios)\n", " sample_rate = config[\"audio\"][\"sample_rate\"]\n", " display(Markdown(f\"{line}\"))\n", - " display(Audio(audio, rate=sample_rate, autoplay=auto_play))\n", + " display(Audio(merged_audio, rate=sample_rate, autoplay=auto_play))\n", "\n", "def denoise(\n", " audio: np.ndarray, bias_spec: np.ndarray, denoiser_strength: float\n",