diff --git a/.github/workflows/api_tests.yml b/.github/workflows/api_tests.yml deleted file mode 100644 index 5a3baaad9e..0000000000 --- a/.github/workflows/api_tests.yml +++ /dev/null @@ -1,53 +0,0 @@ -name: api_tests - -on: - push: - branches: - - main -jobs: - check_skip: - runs-on: ubuntu-latest - if: "! contains(github.event.head_commit.message, '[ci skip]')" - steps: - - run: echo "${{ github.event.head_commit.message }}" - - test: - runs-on: ubuntu-latest - strategy: - fail-fast: false - matrix: - python-version: [3.9, "3.10", "3.11"] - experimental: [false] - steps: - - uses: actions/checkout@v3 - - name: Set up Python ${{ matrix.python-version }} - uses: actions/setup-python@v4 - with: - python-version: ${{ matrix.python-version }} - architecture: x64 - cache: 'pip' - cache-dependency-path: 'requirements*' - - name: check OS - run: cat /etc/os-release - - name: set ENV - run: | - export TRAINER_TELEMETRY=0 - - name: Install dependencies - run: | - sudo apt-get update - sudo apt-get install -y --no-install-recommends git make gcc - sudo apt-get install espeak-ng - make system-deps - - name: Install/upgrade Python setup deps - run: python3 -m pip install --upgrade pip setuptools wheel - - name: Replace scarf urls - run: | - sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json - - name: Install TTS - run: | - python3 -m pip install .[all] - python3 setup.py egg_info - - name: Unit tests - run: make api_tests - env: - COQUI_STUDIO_TOKEN: ${{ secrets.COQUI_STUDIO_TOKEN }} diff --git a/.github/workflows/zoo_tests_tortoise.yml b/.github/workflows/zoo_tests_tortoise.yml deleted file mode 100644 index 31442877e1..0000000000 --- a/.github/workflows/zoo_tests_tortoise.yml +++ /dev/null @@ -1,52 +0,0 @@ -name: zoo-tests-tortoise - -on: - push: - branches: - - main - pull_request: - types: [opened, synchronize, reopened] -jobs: - check_skip: - runs-on: ubuntu-latest - if: "! contains(github.event.head_commit.message, '[ci skip]')" - steps: - - run: echo "${{ github.event.head_commit.message }}" - - test: - runs-on: ubuntu-latest - strategy: - fail-fast: false - matrix: - python-version: [3.9, "3.10", "3.11"] - experimental: [false] - steps: - - uses: actions/checkout@v3 - - name: Set up Python ${{ matrix.python-version }} - uses: actions/setup-python@v4 - with: - python-version: ${{ matrix.python-version }} - architecture: x64 - cache: 'pip' - cache-dependency-path: 'requirements*' - - name: check OS - run: cat /etc/os-release - - name: set ENV - run: export TRAINER_TELEMETRY=0 - - name: Install dependencies - run: | - sudo apt-get update - sudo apt-get install -y git make gcc - sudo apt-get install espeak espeak-ng - make system-deps - - name: Install/upgrade Python setup deps - run: python3 -m pip install --upgrade pip setuptools wheel - - name: Replace scarf urls - run: | - sed -i 's/https:\/\/coqui.gateway.scarf.sh\//https:\/\/github.com\/coqui-ai\/TTS\/releases\/download\//g' TTS/.models.json - - name: Install TTS - run: | - python3 -m pip install .[all] - python3 setup.py egg_info - - name: Unit tests - run: nose2 -F -v -B --with-coverage --coverage TTS tests.zoo_tests.test_models.test_tortoise diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index cae35993dc..ae0ce46048 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -48,7 +48,7 @@ The following steps are tested on an Ubuntu system. 1. Fork 🐸TTS[https://github.com/coqui-ai/TTS] by clicking the fork button at the top right corner of the project page. -2. Clone 🐸TTS and add the main repo as a new remote named ```upsteam```. +2. Clone 🐸TTS and add the main repo as a new remote named ```upstream```. ```bash $ git clone git@github.com:/TTS.git diff --git a/Makefile b/Makefile index 54aa6eeb18..7446848f46 100644 --- a/Makefile +++ b/Makefile @@ -35,9 +35,6 @@ test_zoo: ## run zoo tests. inference_tests: ## run inference tests. nose2 -F -v -B --with-coverage --coverage TTS tests.inference_tests -api_tests: ## run api tests. - nose2 -F -v -B --with-coverage --coverage TTS tests.api_tests - data_tests: ## run data tests. nose2 -F -v -B --with-coverage --coverage TTS tests.data_tests diff --git a/README.md b/README.md index ef16c9b6a1..891118c13d 100644 --- a/README.md +++ b/README.md @@ -7,8 +7,6 @@ - 📣 [🐶Bark](https://github.com/suno-ai/bark) is now available for inference with unconstrained voice cloning. [Docs](https://tts.readthedocs.io/en/dev/models/bark.html) - 📣 You can use [~1100 Fairseq models](https://github.com/facebookresearch/fairseq/tree/main/examples/mms) with 🐸TTS. - 📣 🐸TTS now supports 🐢Tortoise with faster inference. [Docs](https://tts.readthedocs.io/en/dev/models/tortoise.html) -- 📣 **Coqui Studio API** is landed on 🐸TTS. - [Example](https://github.com/coqui-ai/TTS/blob/dev/README.md#-python-api) -- 📣 [**Coqui Studio API**](https://docs.coqui.ai/docs) is live. - 📣 Voice generation with prompts - **Prompt to Voice** - is live on [**Coqui Studio**](https://app.coqui.ai/auth/signin)!! - [Blog Post](https://coqui.ai/blog/tts/prompt-to-voice) - 📣 Voice generation with fusion - **Voice fusion** - is live on [**Coqui Studio**](https://app.coqui.ai/auth/signin). - 📣 Voice cloning is live on [**Coqui Studio**](https://app.coqui.ai/auth/signin). @@ -253,29 +251,6 @@ tts.tts_with_vc_to_file( ) ``` -#### Example using [🐸Coqui Studio](https://coqui.ai) voices. -You access all of your cloned voices and built-in speakers in [🐸Coqui Studio](https://coqui.ai). -To do this, you'll need an API token, which you can obtain from the [account page](https://coqui.ai/account). -After obtaining the API token, you'll need to configure the COQUI_STUDIO_TOKEN environment variable. - -Once you have a valid API token in place, the studio speakers will be displayed as distinct models within the list. -These models will follow the naming convention `coqui_studio/en//coqui_studio` - -```python -# XTTS model -models = TTS(cs_api_model="XTTS").list_models() -# Init TTS with the target studio speaker -tts = TTS(model_name="coqui_studio/en/Torcull Diarmuid/coqui_studio", progress_bar=False) -# Run TTS -tts.tts_to_file(text="This is a test.", language="en", file_path=OUTPUT_PATH) - -# V1 model -models = TTS(cs_api_model="V1").list_models() -# Run TTS with emotion and speed control -# Emotion control only works with V1 model -tts.tts_to_file(text="This is a test.", file_path=OUTPUT_PATH, emotion="Happy", speed=1.5) -``` - #### Example text to speech using **Fairseq models in ~1100 languages** 🤯. For Fairseq models, use the following name format: `tts_models//fairseq/vits`. You can find the language ISO codes [here](https://dl.fbaipublicfiles.com/mms/tts/all-tts-languages.html) @@ -351,12 +326,6 @@ If you don't specify any models, then it uses LJSpeech based English model. $ tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay ``` -- Run TTS and define speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0: - - ``` - $ tts --text "Text for TTS" --model_name "coqui_studio///" --speed 1.2 --out_path output/path/speech.wav - ``` - - Run a TTS model with its default vocoder model: ``` diff --git a/TTS/.models.json b/TTS/.models.json index 1957d78adb..b349e7397b 100644 --- a/TTS/.models.json +++ b/TTS/.models.json @@ -3,12 +3,13 @@ "multilingual": { "multi-dataset": { "xtts_v2": { - "description": "XTTS-v2.0.2 by Coqui with 16 languages.", + "description": "XTTS-v2.0.3 by Coqui with 17 languages.", "hf_url": [ "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/model.pth", "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/config.json", "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/vocab.json", - "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/hash.md5" + "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/hash.md5", + "https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/main/speakers_xtts.pth" ], "model_hash": "10f92b55c512af7a8d39d650547a15a7", "default_vocoder": null, @@ -45,7 +46,7 @@ "hf_url": [ "https://coqui.gateway.scarf.sh/hf/bark/coarse_2.pt", "https://coqui.gateway.scarf.sh/hf/bark/fine_2.pt", - "https://app.coqui.ai/tts_model/text_2.pt", + "https://coqui.gateway.scarf.sh/hf/text_2.pt", "https://coqui.gateway.scarf.sh/hf/bark/config.json", "https://coqui.gateway.scarf.sh/hf/bark/hubert.pt", "https://coqui.gateway.scarf.sh/hf/bark/tokenizer.pth" @@ -270,7 +271,7 @@ "tortoise-v2": { "description": "Tortoise tts model https://github.com/neonbjb/tortoise-tts", "github_rls_url": [ - "https://app.coqui.ai/tts_model/autoregressive.pth", + "https://coqui.gateway.scarf.sh/v0.14.1_models/autoregressive.pth", "https://coqui.gateway.scarf.sh/v0.14.1_models/clvp2.pth", "https://coqui.gateway.scarf.sh/v0.14.1_models/cvvp.pth", "https://coqui.gateway.scarf.sh/v0.14.1_models/diffusion_decoder.pth", diff --git a/TTS/VERSION b/TTS/VERSION index 16eb94e711..2157409059 100644 --- a/TTS/VERSION +++ b/TTS/VERSION @@ -1 +1 @@ -0.21.3 +0.22.0 diff --git a/TTS/api.py b/TTS/api.py index b3aa531b7f..7abc188e74 100644 --- a/TTS/api.py +++ b/TTS/api.py @@ -6,7 +6,6 @@ import numpy as np from torch import nn -from TTS.cs_api import CS_API from TTS.utils.audio.numpy_transforms import save_wav from TTS.utils.manage import ModelManager from TTS.utils.synthesizer import Synthesizer @@ -24,7 +23,6 @@ def __init__( vocoder_path: str = None, vocoder_config_path: str = None, progress_bar: bool = True, - cs_api_model: str = "XTTS", gpu=False, ): """🐸TTS python interface that allows to load and use the released models. @@ -60,9 +58,6 @@ def __init__( vocoder_path (str, optional): Path to the vocoder checkpoint. Defaults to None. vocoder_config_path (str, optional): Path to the vocoder config. Defaults to None. progress_bar (bool, optional): Whether to pring a progress bar while downloading a model. Defaults to True. - cs_api_model (str, optional): Name of the model to use for the Coqui Studio API. Available models are - "XTTS", "V1". You can also use `TTS.cs_api.CS_API" for more control. - Defaults to "XTTS". gpu (bool, optional): Enable/disable GPU. Some models might be too slow on CPU. Defaults to False. """ super().__init__() @@ -70,14 +65,12 @@ def __init__( self.config = load_config(config_path) if config_path else None self.synthesizer = None self.voice_converter = None - self.csapi = None - self.cs_api_model = cs_api_model self.model_name = "" if gpu: warnings.warn("`gpu` will be deprecated. Please use `tts.to(device)` instead.") if model_name is not None and len(model_name) > 0: - if "tts_models" in model_name or "coqui_studio" in model_name: + if "tts_models" in model_name: self.load_tts_model_by_name(model_name, gpu) elif "voice_conversion_models" in model_name: self.load_vc_model_by_name(model_name, gpu) @@ -99,12 +92,6 @@ def is_multi_speaker(self): return self.synthesizer.tts_model.speaker_manager.num_speakers > 1 return False - @property - def is_coqui_studio(self): - if self.model_name is None: - return False - return "coqui_studio" in self.model_name - @property def is_multi_lingual(self): # Not sure what sets this to None, but applied a fix to prevent crashing. @@ -136,14 +123,7 @@ def get_models_file_path(): return Path(__file__).parent / ".models.json" def list_models(self): - try: - csapi = CS_API(model=self.cs_api_model) - models = csapi.list_speakers_as_tts_models() - except ValueError as e: - print(e) - models = [] - manager = ModelManager(models_file=TTS.get_models_file_path(), progress_bar=False, verbose=False) - return manager.list_tts_models() + models + return ModelManager(models_file=TTS.get_models_file_path(), progress_bar=False, verbose=False) def download_model_by_name(self, model_name: str): model_path, config_path, model_item = self.manager.download_model(model_name) @@ -186,30 +166,26 @@ def load_tts_model_by_name(self, model_name: str, gpu: bool = False): TODO: Add tests """ self.synthesizer = None - self.csapi = None self.model_name = model_name - if "coqui_studio" in model_name: - self.csapi = CS_API() - else: - model_path, config_path, vocoder_path, vocoder_config_path, model_dir = self.download_model_by_name( - model_name - ) + model_path, config_path, vocoder_path, vocoder_config_path, model_dir = self.download_model_by_name( + model_name + ) - # init synthesizer - # None values are fetch from the model - self.synthesizer = Synthesizer( - tts_checkpoint=model_path, - tts_config_path=config_path, - tts_speakers_file=None, - tts_languages_file=None, - vocoder_checkpoint=vocoder_path, - vocoder_config=vocoder_config_path, - encoder_checkpoint=None, - encoder_config=None, - model_dir=model_dir, - use_cuda=gpu, - ) + # init synthesizer + # None values are fetch from the model + self.synthesizer = Synthesizer( + tts_checkpoint=model_path, + tts_config_path=config_path, + tts_speakers_file=None, + tts_languages_file=None, + vocoder_checkpoint=vocoder_path, + vocoder_config=vocoder_config_path, + encoder_checkpoint=None, + encoder_config=None, + model_dir=model_dir, + use_cuda=gpu, + ) def load_tts_model_by_path( self, model_path: str, config_path: str, vocoder_path: str = None, vocoder_config: str = None, gpu: bool = False @@ -246,77 +222,17 @@ def _check_arguments( **kwargs, ) -> None: """Check if the arguments are valid for the model.""" - if not self.is_coqui_studio: - # check for the coqui tts models - if self.is_multi_speaker and (speaker is None and speaker_wav is None): - raise ValueError("Model is multi-speaker but no `speaker` is provided.") - if self.is_multi_lingual and language is None: - raise ValueError("Model is multi-lingual but no `language` is provided.") - if not self.is_multi_speaker and speaker is not None and "voice_dir" not in kwargs: - raise ValueError("Model is not multi-speaker but `speaker` is provided.") - if not self.is_multi_lingual and language is not None: - raise ValueError("Model is not multi-lingual but `language` is provided.") - if not emotion is None and not speed is None: - raise ValueError("Emotion and speed can only be used with Coqui Studio models.") - else: - if emotion is None: - emotion = "Neutral" - if speed is None: - speed = 1.0 - # check for the studio models - if speaker_wav is not None: - raise ValueError("Coqui Studio models do not support `speaker_wav` argument.") - if speaker is not None: - raise ValueError("Coqui Studio models do not support `speaker` argument.") - if language is not None and language != "en": - raise ValueError("Coqui Studio models currently support only `language=en` argument.") - if emotion not in ["Neutral", "Happy", "Sad", "Angry", "Dull"]: - raise ValueError(f"Emotion - `{emotion}` - must be one of `Neutral`, `Happy`, `Sad`, `Angry`, `Dull`.") - - def tts_coqui_studio( - self, - text: str, - speaker_name: str = None, - language: str = None, - emotion: str = None, - speed: float = 1.0, - pipe_out=None, - file_path: str = None, - ) -> Union[np.ndarray, str]: - """Convert text to speech using Coqui Studio models. Use `CS_API` class if you are only interested in the API. - - Args: - text (str): - Input text to synthesize. - speaker_name (str, optional): - Speaker name from Coqui Studio. Defaults to None. - language (str): Language of the text. If None, the default language of the speaker is used. Language is only - supported by `XTTS` model. - emotion (str, optional): - Emotion of the speaker. One of "Neutral", "Happy", "Sad", "Angry", "Dull". Emotions are only available - with "V1" model. Defaults to None. - speed (float, optional): - Speed of the speech. Defaults to 1.0. - pipe_out (BytesIO, optional): - Flag to stdout the generated TTS wav file for shell pipe. - file_path (str, optional): - Path to save the output file. When None it returns the `np.ndarray` of waveform. Defaults to None. - - Returns: - Union[np.ndarray, str]: Waveform of the synthesized speech or path to the output file. - """ - speaker_name = self.model_name.split("/")[2] - if file_path is not None: - return self.csapi.tts_to_file( - text=text, - speaker_name=speaker_name, - language=language, - speed=speed, - pipe_out=pipe_out, - emotion=emotion, - file_path=file_path, - )[0] - return self.csapi.tts(text=text, speaker_name=speaker_name, language=language, speed=speed, emotion=emotion)[0] + # check for the coqui tts models + if self.is_multi_speaker and (speaker is None and speaker_wav is None): + raise ValueError("Model is multi-speaker but no `speaker` is provided.") + if self.is_multi_lingual and language is None: + raise ValueError("Model is multi-lingual but no `language` is provided.") + if not self.is_multi_speaker and speaker is not None and "voice_dir" not in kwargs: + raise ValueError("Model is not multi-speaker but `speaker` is provided.") + if not self.is_multi_lingual and language is not None: + raise ValueError("Model is not multi-lingual but `language` is provided.") + if not emotion is None and not speed is None: + raise ValueError("Emotion and speed can only be used with Coqui Studio models. Which is discontinued.") def tts( self, @@ -357,10 +273,6 @@ def tts( self._check_arguments( speaker=speaker, language=language, speaker_wav=speaker_wav, emotion=emotion, speed=speed, **kwargs ) - if self.csapi is not None: - return self.tts_coqui_studio( - text=text, speaker_name=speaker, language=language, emotion=emotion, speed=speed - ) wav = self.synthesizer.tts( text=text, speaker_name=speaker, @@ -419,16 +331,6 @@ def tts_to_file( """ self._check_arguments(speaker=speaker, language=language, speaker_wav=speaker_wav, **kwargs) - if self.csapi is not None: - return self.tts_coqui_studio( - text=text, - speaker_name=speaker, - language=language, - emotion=emotion, - speed=speed, - file_path=file_path, - pipe_out=pipe_out, - ) wav = self.tts( text=text, speaker=speaker, diff --git a/TTS/bin/synthesize.py b/TTS/bin/synthesize.py index d9ec3063e6..b86252ab67 100755 --- a/TTS/bin/synthesize.py +++ b/TTS/bin/synthesize.py @@ -66,12 +66,6 @@ $ tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay ``` -- Run TTS and define speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0: - - ``` - $ tts --text "Text for TTS" --model_name "coqui_studio///" --speed 1.2 --out_path output/path/speech.wav - ``` - - Run a TTS model with its default vocoder model: ``` @@ -222,25 +216,6 @@ def main(): default=None, ) parser.add_argument("--encoder_config_path", type=str, help="Path to speaker encoder config file.", default=None) - - # args for coqui studio - parser.add_argument( - "--cs_model", - type=str, - help="Name of the 🐸Coqui Studio model. Available models are `XTTS`, `V1`.", - ) - parser.add_argument( - "--emotion", - type=str, - help="Emotion to condition the model with. Only available for 🐸Coqui Studio `V1` model.", - default=None, - ) - parser.add_argument( - "--language", - type=str, - help="Language to condition the model with. Only available for 🐸Coqui Studio `XTTS` model.", - default=None, - ) parser.add_argument( "--pipe_out", help="stdout the generated TTS wav file for shell pipe.", @@ -249,13 +224,7 @@ def main(): const=True, default=False, ) - parser.add_argument( - "--speed", - type=float, - help="Speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0.", - default=None, - ) - + # args for multi-speaker synthesis parser.add_argument("--speakers_file_path", type=str, help="JSON file for multi-speaker model.", default=None) parser.add_argument("--language_ids_file_path", type=str, help="JSON file for multi-lingual model.", default=None) @@ -389,7 +358,6 @@ def main(): # CASE1 #list : list pre-trained TTS models if args.list_models: - manager.add_cs_api_models(api.list_models()) manager.list_models() sys.exit() @@ -404,29 +372,7 @@ def main(): manager.model_info_by_full_name(model_query_full_name) sys.exit() - # CASE3: TTS with coqui studio models - if "coqui_studio" in args.model_name: - print(" > Using 🐸Coqui Studio model: ", args.model_name) - api = TTS(model_name=args.model_name, cs_api_model=args.cs_model) - api.tts_to_file( - text=args.text, - emotion=args.emotion, - file_path=args.out_path, - language=args.language, - speed=args.speed, - pipe_out=pipe_out, - ) - print(" > Saving output to ", args.out_path) - return - - if args.language_idx is None and args.language is not None: - msg = ( - "--language is only supported for Coqui Studio models. " - "Use --language_idx to specify the target language for multilingual models." - ) - raise ValueError(msg) - - # CASE4: load pre-trained model paths + # CASE3: load pre-trained model paths if args.model_name is not None and not args.model_path: model_path, config_path, model_item = manager.download_model(args.model_name) # tts model @@ -454,7 +400,7 @@ def main(): if args.vocoder_name is not None and not args.vocoder_path: vocoder_path, vocoder_config_path, _ = manager.download_model(args.vocoder_name) - # CASE5: set custom model paths + # CASE4: set custom model paths if args.model_path is not None: tts_path = args.model_path tts_config_path = args.config_path diff --git a/TTS/config/__init__.py b/TTS/config/__init__.py index 25b4baef81..c5a6dd68e2 100644 --- a/TTS/config/__init__.py +++ b/TTS/config/__init__.py @@ -16,12 +16,9 @@ def read_json_with_comments(json_path): # fallback to json with fsspec.open(json_path, "r", encoding="utf-8") as f: input_str = f.read() - # handle comments - input_str = re.sub(r"\\\n", "", input_str) - input_str = re.sub(r"//.*\n", "\n", input_str) - data = json.loads(input_str) - return data - + # handle comments but not urls with // + input_str = re.sub(r"(\"(?:[^\"\\]|\\.)*\")|(/\*(?:.|[\\n\\r])*?\*/)|(//.*)", lambda m: m.group(1) or m.group(2) or "", input_str) + return json.loads(input_str) def register_config(model_name: str) -> Coqpit: """Find the right config for the given model name. diff --git a/TTS/cs_api.py b/TTS/cs_api.py deleted file mode 100644 index 9dc6c30dd4..0000000000 --- a/TTS/cs_api.py +++ /dev/null @@ -1,317 +0,0 @@ -import http.client -import json -import os -import tempfile -import urllib.request -from typing import Tuple - -import numpy as np -import requests -from scipy.io import wavfile - -from TTS.utils.audio.numpy_transforms import save_wav - - -class Speaker(object): - """Convert dict to object.""" - - def __init__(self, d, is_voice=False): - self.is_voice = is_voice - for k, v in d.items(): - if isinstance(k, (list, tuple)): - setattr(self, k, [Speaker(x) if isinstance(x, dict) else x for x in v]) - else: - setattr(self, k, Speaker(v) if isinstance(v, dict) else v) - - def __repr__(self): - return str(self.__dict__) - - -class CS_API: - """🐸Coqui Studio API Wrapper. - - 🐸Coqui Studio is the most advanced voice generation platform. You can generate new voices by voice cloning, voice - interpolation, or our unique prompt to voice technology. It also provides a set of built-in voices with different - characteristics. You can use these voices to generate new audio files or use them in your applications. - You can use all the built-in and your own 🐸Coqui Studio speakers with this API with an API token. - You can signup to 🐸Coqui Studio from https://app.coqui.ai/auth/signup and get an API token from - https://app.coqui.ai/account. We can either enter the token as an environment variable as - `export COQUI_STUDIO_TOKEN=` or pass it as `CS_API(api_token=)`. - Visit https://app.coqui.ai/api for more information. - - - Args: - api_token (str): 🐸Coqui Studio API token. If not provided, it will be read from the environment variable - `COQUI_STUDIO_TOKEN`. - model (str): 🐸Coqui Studio model. It can be either `V1`, `XTTS`. Default is `XTTS`. - - - Example listing all available speakers: - >>> from TTS.api import CS_API - >>> tts = CS_API() - >>> tts.speakers - - Example listing all emotions: - >>> # emotions are only available for `V1` model - >>> from TTS.api import CS_API - >>> tts = CS_API(model="V1") - >>> tts.emotions - - Example with a built-in 🐸 speaker: - >>> from TTS.api import CS_API - >>> tts = CS_API() - >>> wav, sr = api.tts("Hello world", speaker_name=tts.speakers[0].name) - >>> filepath = tts.tts_to_file(text="Hello world!", speaker_name=tts.speakers[0].name, file_path="output.wav") - - Example with multi-language model: - >>> from TTS.api import CS_API - >>> tts = CS_API(model="XTTS") - >>> wav, sr = api.tts("Hello world", speaker_name=tts.speakers[0].name, language="en") - """ - - MODEL_ENDPOINTS = { - "V1": { - "list_speakers": "https://app.coqui.ai/api/v2/speakers", - "synthesize": "https://app.coqui.ai/api/v2/samples", - "list_voices": "https://app.coqui.ai/api/v2/voices", - }, - "XTTS": { - "list_speakers": "https://app.coqui.ai/api/v2/speakers", - "synthesize": "https://app.coqui.ai/api/v2/samples/xtts/render/", - "list_voices": "https://app.coqui.ai/api/v2/voices/xtts", - }, - } - - SUPPORTED_LANGUAGES = ["en", "es", "de", "fr", "it", "pt", "pl", "tr", "ru", "nl", "cs", "ar", "zh-cn", "ja"] - - def __init__(self, api_token=None, model="XTTS"): - self.api_token = api_token - self.model = model - self.headers = None - self._speakers = None - self._check_token() - - @staticmethod - def ping_api(): - URL = "https://coqui.gateway.scarf.sh/tts/api" - _ = requests.get(URL) - - @property - def speakers(self): - if self._speakers is None: - self._speakers = self.list_all_speakers() - return self._speakers - - @property - def emotions(self): - """Return a list of available emotions. - - TODO: Get this from the API endpoint. - """ - if self.model == "V1": - return ["Neutral", "Happy", "Sad", "Angry", "Dull"] - else: - raise ValueError(f"❗ Emotions are not available for {self.model}.") - - def _check_token(self): - if self.api_token is None: - self.api_token = os.environ.get("COQUI_STUDIO_TOKEN") - self.headers = {"Content-Type": "application/json", "Authorization": f"Bearer {self.api_token}"} - if not self.api_token: - raise ValueError( - "No API token found for 🐸Coqui Studio voices - https://coqui.ai \n" - "Visit 🔗https://app.coqui.ai/account to get one.\n" - "Set it as an environment variable `export COQUI_STUDIO_TOKEN=`\n" - "" - ) - - def list_all_speakers(self): - """Return both built-in Coqui Studio speakers and custom voices created by the user.""" - return self.list_speakers() + self.list_voices() - - def list_speakers(self): - """List built-in Coqui Studio speakers.""" - self._check_token() - conn = http.client.HTTPSConnection("app.coqui.ai") - url = self.MODEL_ENDPOINTS[self.model]["list_speakers"] - conn.request("GET", f"{url}?page=1&per_page=100", headers=self.headers) - res = conn.getresponse() - data = res.read() - return [Speaker(s) for s in json.loads(data)["result"]] - - def list_voices(self): - """List custom voices created by the user.""" - conn = http.client.HTTPSConnection("app.coqui.ai") - url = self.MODEL_ENDPOINTS[self.model]["list_voices"] - conn.request("GET", f"{url}?page=1&per_page=100", headers=self.headers) - res = conn.getresponse() - data = res.read() - return [Speaker(s, True) for s in json.loads(data)["result"]] - - def list_speakers_as_tts_models(self): - """List speakers in ModelManager format.""" - models = [] - for speaker in self.speakers: - model = f"coqui_studio/multilingual/{speaker.name}/{self.model}" - models.append(model) - return models - - def name_to_speaker(self, name): - for speaker in self.speakers: - if speaker.name == name: - return speaker - raise ValueError(f"Speaker {name} not found in {self.speakers}") - - def id_to_speaker(self, speaker_id): - for speaker in self.speakers: - if speaker.id == speaker_id: - return speaker - raise ValueError(f"Speaker {speaker_id} not found.") - - @staticmethod - def url_to_np(url): - tmp_file, _ = urllib.request.urlretrieve(url) - rate, data = wavfile.read(tmp_file) - return data, rate - - @staticmethod - def _create_payload(model, text, speaker, speed, emotion, language): - payload = {} - # if speaker.is_voice: - payload["voice_id"] = speaker.id - # else: - payload["speaker_id"] = speaker.id - - if model == "V1": - payload.update( - { - "emotion": emotion, - "name": speaker.name, - "text": text, - "speed": speed, - } - ) - elif model == "XTTS": - payload.update( - { - "name": speaker.name, - "text": text, - "speed": speed, - "language": language, - } - ) - else: - raise ValueError(f"❗ Unknown model {model}") - return payload - - def _check_tts_args(self, text, speaker_name, speaker_id, emotion, speed, language): - assert text is not None, "❗ text is required for V1 model." - assert speaker_name is not None, "❗ speaker_name is required for V1 model." - if self.model == "V1": - if emotion is None: - emotion = "Neutral" - assert language is None, "❗ language is not supported for V1 model." - elif self.model == "XTTS": - assert emotion is None, f"❗ Emotions are not supported for XTTS model. Use V1 model." - assert language is not None, "❗ Language is required for XTTS model." - assert ( - language in self.SUPPORTED_LANGUAGES - ), f"❗ Language {language} is not yet supported. Check https://docs.coqui.ai/reference/samples_xtts_create." - return text, speaker_name, speaker_id, emotion, speed, language - - def tts( - self, - text: str, - speaker_name: str = None, - speaker_id=None, - emotion=None, - speed=1.0, - language=None, # pylint: disable=unused-argument - ) -> Tuple[np.ndarray, int]: - """Synthesize speech from text. - - Args: - text (str): Text to synthesize. - speaker_name (str): Name of the speaker. You can get the list of speakers with `list_speakers()` and - voices (user generated speakers) with `list_voices()`. - speaker_id (str): Speaker ID. If None, the speaker name is used. - emotion (str): Emotion of the speaker. One of "Neutral", "Happy", "Sad", "Angry", "Dull". Emotions are only - supported by `V1` model. Defaults to None. - speed (float): Speed of the speech. 1.0 is normal speed. - language (str): Language of the text. If None, the default language of the speaker is used. Language is only - supported by `XTTS` model. See https://docs.coqui.ai/reference/samples_xtts_create for supported languages. - """ - self._check_token() - self.ping_api() - - if speaker_name is None and speaker_id is None: - raise ValueError(" [!] Please provide either a `speaker_name` or a `speaker_id`.") - if speaker_id is None: - speaker = self.name_to_speaker(speaker_name) - else: - speaker = self.id_to_speaker(speaker_id) - - text, speaker_name, speaker_id, emotion, speed, language = self._check_tts_args( - text, speaker_name, speaker_id, emotion, speed, language - ) - - conn = http.client.HTTPSConnection("app.coqui.ai") - payload = self._create_payload(self.model, text, speaker, speed, emotion, language) - url = self.MODEL_ENDPOINTS[self.model]["synthesize"] - conn.request("POST", url, json.dumps(payload), self.headers) - res = conn.getresponse() - data = res.read() - try: - wav, sr = self.url_to_np(json.loads(data)["audio_url"]) - except KeyError as e: - raise ValueError(f" [!] 🐸 API returned error: {data}") from e - return wav, sr - - def tts_to_file( - self, - text: str, - speaker_name: str, - speaker_id=None, - emotion=None, - speed=1.0, - pipe_out=None, - language=None, - file_path: str = None, - ) -> str: - """Synthesize speech from text and save it to a file. - - Args: - text (str): Text to synthesize. - speaker_name (str): Name of the speaker. You can get the list of speakers with `list_speakers()` and - voices (user generated speakers) with `list_voices()`. - speaker_id (str): Speaker ID. If None, the speaker name is used. - emotion (str): Emotion of the speaker. One of "Neutral", "Happy", "Sad", "Angry", "Dull". - speed (float): Speed of the speech. 1.0 is normal speed. - pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe. - language (str): Language of the text. If None, the default language of the speaker is used. Language is only - supported by `XTTS` model. Currently supports en, de, es, fr, it, pt, pl. Defaults to "en". - file_path (str): Path to save the file. If None, a temporary file is created. - """ - if file_path is None: - file_path = tempfile.mktemp(".wav") - wav, sr = self.tts(text, speaker_name, speaker_id, emotion, speed, language) - save_wav(wav=wav, path=file_path, sample_rate=sr, pipe_out=pipe_out) - return file_path - - -if __name__ == "__main__": - import time - - api = CS_API() - print(api.speakers) - print(api.list_speakers_as_tts_models()) - - ts = time.time() - wav, sr = api.tts( - "It took me quite a long time to develop a voice.", language="en", speaker_name=api.speakers[0].name - ) - print(f" [i] XTTS took {time.time() - ts:.2f}s") - - filepath = api.tts_to_file( - text="Hello world!", speaker_name=api.speakers[0].name, language="en", file_path="output.wav" - ) diff --git a/TTS/tts/layers/xtts/trainer/gpt_trainer.py b/TTS/tts/layers/xtts/trainer/gpt_trainer.py index 6276f60af6..9a7a1d7783 100644 --- a/TTS/tts/layers/xtts/trainer/gpt_trainer.py +++ b/TTS/tts/layers/xtts/trainer/gpt_trainer.py @@ -321,7 +321,10 @@ def eval_step(self, batch, criterion): def on_train_epoch_start(self, trainer): trainer.model.eval() # the whole model to eval # put gpt model in training mode - trainer.model.xtts.gpt.train() + if hasattr(trainer.model, "module") and hasattr(trainer.model.module, "xtts"): + trainer.model.module.xtts.gpt.train() + else: + trainer.model.xtts.gpt.train() def on_init_end(self, trainer): # pylint: disable=W0613 # ignore similarities.pth on clearml save/upload @@ -387,7 +390,8 @@ def get_data_loader( else: loader = DataLoader( dataset, - batch_sampler=sampler, + sampler=sampler, + batch_size = config.eval_batch_size if is_eval else config.batch_size, collate_fn=dataset.collate_fn, num_workers=config.num_eval_loader_workers if is_eval else config.num_loader_workers, pin_memory=False, diff --git a/TTS/tts/layers/xtts/xtts_manager.py b/TTS/tts/layers/xtts/xtts_manager.py new file mode 100644 index 0000000000..3e7d0f6c91 --- /dev/null +++ b/TTS/tts/layers/xtts/xtts_manager.py @@ -0,0 +1,34 @@ +import torch + +class SpeakerManager(): + def __init__(self, speaker_file_path=None): + self.speakers = torch.load(speaker_file_path) + + @property + def name_to_id(self): + return self.speakers.keys() + + @property + def num_speakers(self): + return len(self.name_to_id) + + @property + def speaker_names(self): + return list(self.name_to_id.keys()) + + +class LanguageManager(): + def __init__(self, config): + self.langs = config["languages"] + + @property + def name_to_id(self): + return self.langs + + @property + def num_languages(self): + return len(self.name_to_id) + + @property + def language_names(self): + return list(self.name_to_id) diff --git a/TTS/tts/layers/xtts/zh_num2words.py b/TTS/tts/layers/xtts/zh_num2words.py index ea6d98d3da..e59ccb6630 100644 --- a/TTS/tts/layers/xtts/zh_num2words.py +++ b/TTS/tts/layers/xtts/zh_num2words.py @@ -65,7 +65,7 @@ CN_PUNCS = CN_PUNCS_STOP + CN_PUNCS_NONSTOP PUNCS = CN_PUNCS + string.punctuation -PUNCS_TRANSFORM = str.maketrans(PUNCS, " " * len(PUNCS), "") # replace puncs with space +PUNCS_TRANSFORM = str.maketrans(PUNCS, "," * len(PUNCS), "") # replace puncs with English comma # https://zh.wikipedia.org/wiki/全行和半行 diff --git a/TTS/tts/models/forward_tts.py b/TTS/tts/models/forward_tts.py index 9e1b1c4097..b6e9ac8a14 100644 --- a/TTS/tts/models/forward_tts.py +++ b/TTS/tts/models/forward_tts.py @@ -241,7 +241,7 @@ def __init__( ) self.duration_predictor = DurationPredictor( - self.args.hidden_channels + self.embedded_speaker_dim, + self.args.hidden_channels, self.args.duration_predictor_hidden_channels, self.args.duration_predictor_kernel_size, self.args.duration_predictor_dropout_p, @@ -249,7 +249,7 @@ def __init__( if self.args.use_pitch: self.pitch_predictor = DurationPredictor( - self.args.hidden_channels + self.embedded_speaker_dim, + self.args.hidden_channels, self.args.pitch_predictor_hidden_channels, self.args.pitch_predictor_kernel_size, self.args.pitch_predictor_dropout_p, @@ -263,7 +263,7 @@ def __init__( if self.args.use_energy: self.energy_predictor = DurationPredictor( - self.args.hidden_channels + self.embedded_speaker_dim, + self.args.hidden_channels, self.args.energy_predictor_hidden_channels, self.args.energy_predictor_kernel_size, self.args.energy_predictor_dropout_p, @@ -299,7 +299,8 @@ def init_multispeaker(self, config: Coqpit): if config.use_d_vector_file: self.embedded_speaker_dim = config.d_vector_dim if self.args.d_vector_dim != self.args.hidden_channels: - self.proj_g = nn.Conv1d(self.args.d_vector_dim, self.args.hidden_channels, 1) + #self.proj_g = nn.Conv1d(self.args.d_vector_dim, self.args.hidden_channels, 1) + self.proj_g = nn.Linear(in_features=self.args.d_vector_dim, out_features=self.args.hidden_channels) # init speaker embedding layer if config.use_speaker_embedding and not config.use_d_vector_file: print(" > Init speaker_embedding layer.") @@ -403,10 +404,13 @@ def _forward_encoder( # [B, T, C] x_emb = self.emb(x) # encoder pass - o_en = self.encoder(torch.transpose(x_emb, 1, -1), x_mask) + #o_en = self.encoder(torch.transpose(x_emb, 1, -1), x_mask) + o_en = self.encoder(torch.transpose(x_emb, 1, -1), x_mask, g) # speaker conditioning # TODO: try different ways of conditioning - if g is not None: + if g is not None: + if hasattr(self, "proj_g"): + g = self.proj_g(g.view(g.shape[0], -1)).unsqueeze(-1) o_en = o_en + g return o_en, x_mask, g, x_emb diff --git a/TTS/tts/models/xtts.py b/TTS/tts/models/xtts.py index 6b8cc59101..83812f377f 100644 --- a/TTS/tts/models/xtts.py +++ b/TTS/tts/models/xtts.py @@ -11,6 +11,7 @@ from TTS.tts.layers.xtts.hifigan_decoder import HifiDecoder from TTS.tts.layers.xtts.stream_generator import init_stream_support from TTS.tts.layers.xtts.tokenizer import VoiceBpeTokenizer, split_sentence +from TTS.tts.layers.xtts.xtts_manager import SpeakerManager, LanguageManager from TTS.tts.models.base_tts import BaseTTS from TTS.utils.io import load_fsspec @@ -378,7 +379,7 @@ def get_conditioning_latents( return gpt_cond_latents, speaker_embedding - def synthesize(self, text, config, speaker_wav, language, **kwargs): + def synthesize(self, text, config, speaker_wav, language, speaker_id=None, **kwargs): """Synthesize speech with the given input text. Args: @@ -393,12 +394,6 @@ def synthesize(self, text, config, speaker_wav, language, **kwargs): `text_input` as text token IDs after tokenizer, `voice_samples` as samples used for cloning, `conditioning_latents` as latents used at inference. - """ - return self.inference_with_config(text, config, ref_audio_path=speaker_wav, language=language, **kwargs) - - def inference_with_config(self, text, config, ref_audio_path, language, **kwargs): - """ - inference with config """ assert ( "zh-cn" if language == "zh" else language in self.config.languages @@ -410,13 +405,18 @@ def inference_with_config(self, text, config, ref_audio_path, language, **kwargs "repetition_penalty": config.repetition_penalty, "top_k": config.top_k, "top_p": config.top_p, + } + settings.update(kwargs) # allow overriding of preset settings with kwargs + if speaker_id is not None: + gpt_cond_latent, speaker_embedding = self.speaker_manager.speakers[speaker_id].values() + return self.inference(text, language, gpt_cond_latent, speaker_embedding, **settings) + settings.update({ "gpt_cond_len": config.gpt_cond_len, "gpt_cond_chunk_len": config.gpt_cond_chunk_len, "max_ref_len": config.max_ref_len, "sound_norm_refs": config.sound_norm_refs, - } - settings.update(kwargs) # allow overriding of preset settings with kwargs - return self.full_inference(text, ref_audio_path, language, **settings) + }) + return self.full_inference(text, speaker_wav, language, **settings) @torch.inference_mode() def full_inference( @@ -520,6 +520,8 @@ def inference( ): language = language.split("-")[0] # remove the country code length_scale = 1.0 / max(speed, 0.05) + gpt_cond_latent = gpt_cond_latent.to(self.device) + speaker_embedding = speaker_embedding.to(self.device) if enable_text_splitting: text = split_sentence(text, language, self.tokenizer.char_limits[language]) else: @@ -628,6 +630,8 @@ def inference_stream( ): language = language.split("-")[0] # remove the country code length_scale = 1.0 / max(speed, 0.05) + gpt_cond_latent = gpt_cond_latent.to(self.device) + speaker_embedding = speaker_embedding.to(self.device) if enable_text_splitting: text = split_sentence(text, language, self.tokenizer.char_limits[language]) else: @@ -733,6 +737,7 @@ def load_checkpoint( eval=True, strict=True, use_deepspeed=False, + speaker_file_path=None, ): """ Loads a checkpoint from disk and initializes the model's state and tokenizer. @@ -751,6 +756,12 @@ def load_checkpoint( model_path = checkpoint_path or os.path.join(checkpoint_dir, "model.pth") vocab_path = vocab_path or os.path.join(checkpoint_dir, "vocab.json") + speaker_file_path = speaker_file_path or os.path.join(checkpoint_dir, "speakers_xtts.pth") + + self.language_manager = LanguageManager(config) + self.speaker_manager = None + if os.path.exists(speaker_file_path): + self.speaker_manager = SpeakerManager(speaker_file_path) if os.path.exists(vocab_path): self.tokenizer = VoiceBpeTokenizer(vocab_file=vocab_path) diff --git a/TTS/utils/manage.py b/TTS/utils/manage.py index 3952504d0b..3a527f4609 100644 --- a/TTS/utils/manage.py +++ b/TTS/utils/manage.py @@ -11,7 +11,7 @@ import requests from tqdm import tqdm -from TTS.config import load_config +from TTS.config import load_config, read_json_with_comments from TTS.utils.generic_utils import get_user_data_dir LICENSE_URLS = { @@ -65,30 +65,7 @@ def read_models_file(self, file_path): Args: file_path (str): path to .models.json. """ - with open(file_path, "r", encoding="utf-8") as json_file: - self.models_dict = json.load(json_file) - - def add_cs_api_models(self, model_list: List[str]): - """Add list of Coqui Studio model names that are returned from the api - - Each has the following format `/en//` - """ - - def _add_model(model_name: str): - if not "coqui_studio" in model_name: - return - model_type, lang, dataset, model = model_name.split("/") - if model_type not in self.models_dict: - self.models_dict[model_type] = {} - if lang not in self.models_dict[model_type]: - self.models_dict[model_type][lang] = {} - if dataset not in self.models_dict[model_type][lang]: - self.models_dict[model_type][lang][dataset] = {} - if model not in self.models_dict[model_type][lang][dataset]: - self.models_dict[model_type][lang][dataset][model] = {} - - for model_name in model_list: - _add_model(model_name) + self.models_dict = read_json_with_comments(file_path) def _list_models(self, model_type, model_count=0): if self.verbose: @@ -315,6 +292,7 @@ def _set_model_item(self, model_name): f"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/{model_version}/config.json", f"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/{model_version}/vocab.json", f"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/{model_version}/hash.md5", + f"https://coqui.gateway.scarf.sh/hf-coqui/XTTS-v2/{model_version}/speakers_xtts.pth", ], } else: @@ -332,9 +310,9 @@ def _set_model_item(self, model_name): def ask_tos(model_full_path): """Ask the user to agree to the terms of service""" tos_path = os.path.join(model_full_path, "tos_agreed.txt") - print(" > You must agree to the terms of service to use this model.") - print(" | > Please see the terms of service at https://coqui.ai/cpml.txt") - print(' | > "I have read, understood and agreed to the Terms and Conditions." - [y/n]') + print(" > You must confirm the following:") + print(' | > "I have purchased a commercial license from Coqui: licensing@coqui.ai"') + print(' | > "Otherwise, I agree to the terms of the non-commercial CPML: https://coqui.ai/cpml" - [y/n]') answer = input(" | | > ") if answer.lower() == "y": with open(tos_path, "w", encoding="utf-8") as f: diff --git a/TTS/utils/synthesizer.py b/TTS/utils/synthesizer.py index 781561f973..b98647c30c 100644 --- a/TTS/utils/synthesizer.py +++ b/TTS/utils/synthesizer.py @@ -305,7 +305,7 @@ def tts( speaker_embedding = None speaker_id = None if self.tts_speakers_file or hasattr(self.tts_model.speaker_manager, "name_to_id"): - if speaker_name and isinstance(speaker_name, str): + if speaker_name and isinstance(speaker_name, str) and not self.tts_config.model == "xtts": if self.tts_config.use_d_vector_file: # get the average speaker embedding from the saved d_vectors. speaker_embedding = self.tts_model.speaker_manager.get_mean_embedding( @@ -335,7 +335,9 @@ def tts( # handle multi-lingual language_id = None if self.tts_languages_file or ( - hasattr(self.tts_model, "language_manager") and self.tts_model.language_manager is not None + hasattr(self.tts_model, "language_manager") + and self.tts_model.language_manager is not None + and not self.tts_config.model == "xtts" ): if len(self.tts_model.language_manager.name_to_id) == 1: language_id = list(self.tts_model.language_manager.name_to_id.values())[0] @@ -366,6 +368,7 @@ def tts( if ( speaker_wav is not None and self.tts_model.speaker_manager is not None + and hasattr(self.tts_model.speaker_manager, "encoder_ap") and self.tts_model.speaker_manager.encoder_ap is not None ): speaker_embedding = self.tts_model.speaker_manager.compute_embedding_from_clip(speaker_wav) diff --git a/docs/source/configuration.md b/docs/source/configuration.md index cde7e073e9..ada61e16db 100644 --- a/docs/source/configuration.md +++ b/docs/source/configuration.md @@ -56,4 +56,4 @@ ModelConfig() In the example above, ```ModelConfig()``` is the final configuration that the model receives and it has all the fields necessary for the model. -We host pre-defined model configurations under ```TTS//configs/```.Although we recommend a unified config class, you can decompose it as you like as for your custom models as long as all the fields for the trainer, model, and inference APIs are provided. \ No newline at end of file +We host pre-defined model configurations under ```TTS//configs/```. Although we recommend a unified config class, you can decompose it as you like as for your custom models as long as all the fields for the trainer, model, and inference APIs are provided. diff --git a/docs/source/finetuning.md b/docs/source/finetuning.md index c236260d0c..069f565137 100644 --- a/docs/source/finetuning.md +++ b/docs/source/finetuning.md @@ -21,7 +21,7 @@ them and fine-tune it for your own dataset. This will help you in two main ways: Fine-tuning comes to the rescue in this case. You can take one of our pre-trained models and fine-tune it on your own speech dataset and achieve reasonable results with only a couple of hours of data. - However, note that, fine-tuning does not ensure great results. The model performance is still depends on the + However, note that, fine-tuning does not ensure great results. The model performance still depends on the {ref}`dataset quality ` and the hyper-parameters you choose for fine-tuning. Therefore, it still takes a bit of tinkering. @@ -41,7 +41,7 @@ them and fine-tune it for your own dataset. This will help you in two main ways: tts --list_models ``` - The command above lists the the models in a naming format as ```///```. + The command above lists the models in a naming format as ```///```. Or you can manually check the `.model.json` file in the project directory. diff --git a/docs/source/formatting_your_dataset.md b/docs/source/formatting_your_dataset.md index 796c7b6d06..23c497d0bf 100644 --- a/docs/source/formatting_your_dataset.md +++ b/docs/source/formatting_your_dataset.md @@ -7,7 +7,7 @@ If you have a single audio file and you need to split it into clips, there are d It is also important to use a lossless audio file format to prevent compression artifacts. We recommend using `wav` file format. -Let's assume you created the audio clips and their transcription. You can collect all your clips under a folder. Let's call this folder `wavs`. +Let's assume you created the audio clips and their transcription. You can collect all your clips in a folder. Let's call this folder `wavs`. ``` /wavs @@ -17,7 +17,7 @@ Let's assume you created the audio clips and their transcription. You can collec ... ``` -You can either create separate transcription files for each clip or create a text file that maps each audio clip to its transcription. In this file, each column must be delimitered by a special character separating the audio file name, the transcription and the normalized transcription. And make sure that the delimiter is not used in the transcription text. +You can either create separate transcription files for each clip or create a text file that maps each audio clip to its transcription. In this file, each column must be delimited by a special character separating the audio file name, the transcription and the normalized transcription. And make sure that the delimiter is not used in the transcription text. We recommend the following format delimited by `|`. In the following example, `audio1`, `audio2` refer to files `audio1.wav`, `audio2.wav` etc. @@ -55,7 +55,7 @@ For more info about dataset qualities and properties check our [post](https://gi After you collect and format your dataset, you need to check two things. Whether you need a `formatter` and a `text_cleaner`. The `formatter` loads the text file (created above) as a list and the `text_cleaner` performs a sequence of text normalization operations that converts the raw text into the spoken representation (e.g. converting numbers to text, acronyms, and symbols to the spoken format). -If you use a different dataset format then the LJSpeech or the other public datasets that 🐸TTS supports, then you need to write your own `formatter`. +If you use a different dataset format than the LJSpeech or the other public datasets that 🐸TTS supports, then you need to write your own `formatter`. If your dataset is in a new language or it needs special normalization steps, then you need a new `text_cleaner`. diff --git a/docs/source/implementing_a_new_language_frontend.md b/docs/source/implementing_a_new_language_frontend.md index f4f6a04a5f..2041352d64 100644 --- a/docs/source/implementing_a_new_language_frontend.md +++ b/docs/source/implementing_a_new_language_frontend.md @@ -2,11 +2,11 @@ - Language frontends are located under `TTS.tts.utils.text` - Each special language has a separate folder. -- Each folder containst all the utilities for processing the text input. +- Each folder contains all the utilities for processing the text input. - `TTS.tts.utils.text.phonemizers` contains the main phonemizer for a language. This is the class that uses the utilities from the previous step and used to convert the text to phonemes or graphemes for the model. - After you implement your phonemizer, you need to add it to the `TTS/tts/utils/text/phonemizers/__init__.py` to be able to map the language code in the model config - `config.phoneme_language` - to the phonemizer class and initiate the phonemizer automatically. - You should also add tests to `tests/text_tests` if you want to make a PR. -We suggest you to check the available implementations as reference. Good luck! \ No newline at end of file +We suggest you to check the available implementations as reference. Good luck! diff --git a/docs/source/implementing_a_new_model.md b/docs/source/implementing_a_new_model.md index e2a0437e9a..1bf7a8822e 100644 --- a/docs/source/implementing_a_new_model.md +++ b/docs/source/implementing_a_new_model.md @@ -145,7 +145,7 @@ class MyModel(BaseTTS): Args: ap (AudioProcessor): audio processor used at training. batch (Dict): Model inputs used at the previous training step. - outputs (Dict): Model outputs generated at the previoud training step. + outputs (Dict): Model outputs generated at the previous training step. Returns: Tuple[Dict, np.ndarray]: training plots and output waveform. @@ -183,7 +183,7 @@ class MyModel(BaseTTS): ... def get_optimizer(self) -> Union["Optimizer", List["Optimizer"]]: - """Setup an return optimizer or optimizers.""" + """Setup a return optimizer or optimizers.""" pass def get_lr(self) -> Union[float, List[float]]: diff --git a/docs/source/inference.md b/docs/source/inference.md index 611a2445bf..56bccfb5b2 100644 --- a/docs/source/inference.md +++ b/docs/source/inference.md @@ -172,48 +172,6 @@ tts.tts_with_vc_to_file( ) ``` -#### Example text to speech using [🐸Coqui Studio](https://coqui.ai) models. - -You can use all of your available speakers in the studio. -[🐸Coqui Studio](https://coqui.ai) API token is required. You can get it from the [account page](https://coqui.ai/account). -You should set the `COQUI_STUDIO_TOKEN` environment variable to use the API token. - -```python -# If you have a valid API token set you will see the studio speakers as separate models in the list. -# The name format is coqui_studio/en//coqui_studio -models = TTS().list_models() -# Init TTS with the target studio speaker -tts = TTS(model_name="coqui_studio/en/Torcull Diarmuid/coqui_studio", progress_bar=False) -# Run TTS -tts.tts_to_file(text="This is a test.", file_path=OUTPUT_PATH) -# Run TTS with emotion and speed control -tts.tts_to_file(text="This is a test.", file_path=OUTPUT_PATH, emotion="Happy", speed=1.5) -``` - -If you just need 🐸 Coqui Studio speakers, you can use `CS_API`. It is a wrapper around the 🐸 Coqui Studio API. - -```python -from TTS.api import CS_API - -# Init 🐸 Coqui Studio API -# you can either set the API token as an environment variable `COQUI_STUDIO_TOKEN` or pass it as an argument. - -# XTTS - Best quality and life-like speech in multiple languages. See https://docs.coqui.ai/reference/samples_xtts_create for supported languages. -api = CS_API(api_token=, model="XTTS") -api.speakers # all the speakers are available with all the models. -api.list_speakers() -api.list_voices() -wav, sample_rate = api.tts(text="This is a test.", speaker=api.speakers[0].name, emotion="Happy", language="en", speed=1.5) - -# V1 - Fast and lightweight TTS in EN with emotion control. -api = CS_API(api_token=, model="V1") -api.speakers -api.emotions # emotions are only for the V1 model. -api.list_speakers() -api.list_voices() -wav, sample_rate = api.tts(text="This is a test.", speaker=api.speakers[0].name, emotion="Happy", speed=1.5) -``` - #### Example text to speech using **Fairseq models in ~1100 languages** 🤯. For these models use the following name format: `tts_models//fairseq/vits`. diff --git a/docs/source/marytts.md b/docs/source/marytts.md index 81d547107d..9091ca330f 100644 --- a/docs/source/marytts.md +++ b/docs/source/marytts.md @@ -2,13 +2,13 @@ ## What is Mary-TTS? -[Mary (Modular Architecture for Research in sYynthesis) Text-to-Speech](http://mary.dfki.de/) is an open-source (GNU LGPL license), multilingual Text-to-Speech Synthesis platform written in Java. It was originally developed as a collaborative project of [DFKI’s](http://www.dfki.de/web) Language Technology Lab and the [Institute of Phonetics](http://www.coli.uni-saarland.de/groups/WB/Phonetics/) at Saarland University, Germany. It is now maintained by the Multimodal Speech Processing Group in the [Cluster of Excellence MMCI](https://www.mmci.uni-saarland.de/) and DFKI. +[Mary (Modular Architecture for Research in sYnthesis) Text-to-Speech](http://mary.dfki.de/) is an open-source (GNU LGPL license), multilingual Text-to-Speech Synthesis platform written in Java. It was originally developed as a collaborative project of [DFKI’s](http://www.dfki.de/web) Language Technology Lab and the [Institute of Phonetics](http://www.coli.uni-saarland.de/groups/WB/Phonetics/) at Saarland University, Germany. It is now maintained by the Multimodal Speech Processing Group in the [Cluster of Excellence MMCI](https://www.mmci.uni-saarland.de/) and DFKI. MaryTTS has been around for a very! long time. Version 3.0 even dates back to 2006, long before Deep Learning was a broadly known term and the last official release was version 5.2 in 2016. You can check out this OpenVoice-Tech page to learn more: https://openvoice-tech.net/index.php/MaryTTS ## Why Mary-TTS compatibility is relevant -Due to it's open-source nature, relatively high quality voices and fast synthetization speed Mary-TTS was a popular choice in the past and many tools implemented API support over the years like screen-readers (NVDA + SpeechHub), smart-home HUBs (openHAB, Home Assistant) or voice assistants (Rhasspy, Mycroft, SEPIA). A compatibility layer for Coqui-TTS will ensure that these tools can use Coqui as a drop-in replacement and get even better voices right away. +Due to its open-source nature, relatively high quality voices and fast synthetization speed Mary-TTS was a popular choice in the past and many tools implemented API support over the years like screen-readers (NVDA + SpeechHub), smart-home HUBs (openHAB, Home Assistant) or voice assistants (Rhasspy, Mycroft, SEPIA). A compatibility layer for Coqui-TTS will ensure that these tools can use Coqui as a drop-in replacement and get even better voices right away. ## API and code examples @@ -40,4 +40,4 @@ You can enter the same URLs in your browser and check-out the results there as w ### How it works and limitations A classic Mary-TTS server would usually show all installed locales and voices via the corresponding endpoints and accept the parameters `LOCALE` and `VOICE` for processing. For Coqui-TTS we usually start the server with one specific locale and model and thus cannot return all available options. Instead we return the active locale and use the model name as "voice". Since we only have one active model and always want to return a WAV-file, we currently ignore all other processing parameters except `INPUT_TEXT`. Since the gender is not defined for models in Coqui-TTS we always return `u` (undefined). -We think that this is an acceptable compromise, since users are often only interested in one specific voice anyways, but the API might get extended in the future to support multiple languages and voices at the same time. \ No newline at end of file +We think that this is an acceptable compromise, since users are often only interested in one specific voice anyways, but the API might get extended in the future to support multiple languages and voices at the same time. diff --git a/docs/source/models/tortoise.md b/docs/source/models/tortoise.md index 2df6da7649..1a8e9ca8e9 100644 --- a/docs/source/models/tortoise.md +++ b/docs/source/models/tortoise.md @@ -1,6 +1,6 @@ # 🐢 Tortoise Tortoise is a very expressive TTS system with impressive voice cloning capabilities. It is based on an GPT like autogressive acoustic model that converts input -text to discritized acouistic tokens, a diffusion model that converts these tokens to melspeectrogram frames and a Univnet vocoder to convert the spectrograms to +text to discritized acoustic tokens, a diffusion model that converts these tokens to melspectrogram frames and a Univnet vocoder to convert the spectrograms to the final audio signal. The important downside is that Tortoise is very slow compared to the parallel TTS models like VITS. Big thanks to 👑[@manmay-nakhashi](https://github.com/manmay-nakhashi) who helped us implement Tortoise in 🐸TTS. diff --git a/docs/source/models/xtts.md b/docs/source/models/xtts.md index e5da50ff4e..b979d04f6e 100644 --- a/docs/source/models/xtts.md +++ b/docs/source/models/xtts.md @@ -21,7 +21,7 @@ a few tricks to make it faster and support streaming inference. - Across the board quality improvements. ### Code -Current implementation only supports inference. +Current implementation only supports inference and GPT encoder training. ### Languages As of now, XTTS-v2 supports 16 languages: English (en), Spanish (es), French (fr), German (de), Italian (it), Portuguese (pt), Polish (pl), Turkish (tr), Russian (ru), Dutch (nl), Czech (cs), Arabic (ar), Chinese (zh-cn), Japanese (ja), Hungarian (hu) and Korean (ko). @@ -36,9 +36,71 @@ Come and join in our 🐸Community. We're active on [Discord](https://discord.gg You can also mail us at info@coqui.ai. ### Inference + +#### 🐸TTS Command line + +You can check all supported languages with the following command: + +```console + tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ + --list_language_idx +``` + +You can check all Coqui available speakers with the following command: + +```console + tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ + --list_speaker_idx +``` + +##### Coqui speakers +You can do inference using one of the available speakers using the following command: + +```console + tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ + --text "It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent." \ + --speaker_idx "Ana Florence" \ + --language_idx en \ + --use_cuda true +``` + +##### Clone a voice +You can clone a speaker voice using a single or multiple references: + +###### Single reference + +```console + tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ + --text "Bugün okula gitmek istemiyorum." \ + --speaker_wav /path/to/target/speaker.wav \ + --language_idx tr \ + --use_cuda true +``` + +###### Multiple references +```console + tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ + --text "Bugün okula gitmek istemiyorum." \ + --speaker_wav /path/to/target/speaker.wav /path/to/target/speaker_2.wav /path/to/target/speaker_3.wav \ + --language_idx tr \ + --use_cuda true +``` +or for all wav files in a directory you can use: + +```console + tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ + --text "Bugün okula gitmek istemiyorum." \ + --speaker_wav /path/to/target/*.wav \ + --language_idx tr \ + --use_cuda true +``` + #### 🐸TTS API -##### Single reference +##### Clone a voice +You can clone a speaker voice using a single or multiple references: + +###### Single reference Splits the text into sentences and generates audio for each sentence. The audio files are then concatenated to produce the final audio. You can optionally disable sentence splitting for better coherence but more VRAM and possibly hitting models context length limit. @@ -56,7 +118,7 @@ tts.tts_to_file(text="It took me quite a long time to develop a voice, and now t ) ``` -##### Multiple references +###### Multiple references You can pass multiple audio files to the `speaker_wav` argument for better voice cloning. @@ -81,82 +143,53 @@ tts.tts_to_file(text="It took me quite a long time to develop a voice, and now t language="en") ``` -##### Streaming inference +##### Coqui speakers -XTTS supports streaming inference. This is useful for real-time applications. +You can do inference using one of the available speakers using the following code: ```python -import os -import time -import torch -import torchaudio - -print("Loading model...") +from TTS.api import TTS tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2", gpu=True) -model = tts.synthesizer.tts_model -print("Computing speaker latents...") -gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(audio_path=["reference.wav"]) +# generate speech by cloning a voice using default settings +tts.tts_to_file(text="It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.", + file_path="output.wav", + speaker="Ana Florence", + language="en", + split_sentences=True + ) +``` -print("Inference...") -t0 = time.time() -stream_generator = model.inference_stream( - "It took me quite a long time to develop a voice and now that I have it I am not going to be silent.", - "en", - gpt_cond_latent, - speaker_embedding -) -wav_chuncks = [] -for i, chunk in enumerate(stream_generator): - if i == 0: - print(f"Time to first chunck: {time.time() - t0}") - print(f"Received chunk {i} of audio length {chunk.shape[-1]}") - wav_chuncks.append(chunk) -wav = torch.cat(wav_chuncks, dim=0) -torchaudio.save("xtts_streaming.wav", wav.squeeze().unsqueeze(0).cpu(), 24000) -``` +#### 🐸TTS Model API -#### 🐸TTS Command line +To use the model API, you need to download the model files and pass config and model file paths manually. -##### Single reference -```console - tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ - --text "Bugün okula gitmek istemiyorum." \ - --speaker_wav /path/to/target/speaker.wav \ - --language_idx tr \ - --use_cuda true -``` +#### Manual Inference -##### Multiple references -```console - tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ - --text "Bugün okula gitmek istemiyorum." \ - --speaker_wav /path/to/target/speaker.wav /path/to/target/speaker_2.wav /path/to/target/speaker_3.wav \ - --language_idx tr \ - --use_cuda true -``` -or for all wav files in a directory you can use: +If you want to be able to `load_checkpoint` with `use_deepspeed=True` and **enjoy the speedup**, you need to install deepspeed first. ```console - tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \ - --text "Bugün okula gitmek istemiyorum." \ - --speaker_wav /path/to/target/*.wav \ - --language_idx tr \ - --use_cuda true +pip install deepspeed==0.10.3 ``` -#### 🐸TTS Model API +##### inference parameters -To use the model API, you need to download the model files and pass config and model file paths manually. +- `text`: The text to be synthesized. +- `language`: The language of the text to be synthesized. +- `gpt_cond_latent`: The latent vector you get with get_conditioning_latents. (You can cache for faster inference with same speaker) +- `speaker_embedding`: The speaker embedding you get with get_conditioning_latents. (You can cache for faster inference with same speaker) +- `temperature`: The softmax temperature of the autoregressive model. Defaults to 0.65. +- `length_penalty`: A length penalty applied to the autoregressive decoder. Higher settings causes the model to produce more terse outputs. Defaults to 1.0. +- `repetition_penalty`: A penalty that prevents the autoregressive decoder from repeating itself during decoding. Can be used to reduce the incidence of long silences or "uhhhhhhs", etc. Defaults to 2.0. +- `top_k`: Lower values mean the decoder produces more "likely" (aka boring) outputs. Defaults to 50. +- `top_p`: Lower values mean the decoder produces more "likely" (aka boring) outputs. Defaults to 0.8. +- `speed`: The speed rate of the generated audio. Defaults to 1.0. (can produce artifacts if far from 1.0) +- `enable_text_splitting`: Whether to split the text into sentences and generate audio for each sentence. It allows you to have infinite input length but might loose important context between sentences. Defaults to True. -##### Calling manually -If you want to be able to run with `use_deepspeed=True` and **enjoy the speedup**, you need to install deepspeed first. +##### Inference -```console -pip install deepspeed==0.10.3 -``` ```python import os diff --git a/tests/api_tests/__init__.py b/tests/api_tests/__init__.py deleted file mode 100644 index e69de29bb2..0000000000 diff --git a/tests/api_tests/test_python_api.py b/tests/api_tests/test_python_api.py deleted file mode 100644 index 2025fcd9c6..0000000000 --- a/tests/api_tests/test_python_api.py +++ /dev/null @@ -1,113 +0,0 @@ -import os -import unittest - -from tests import get_tests_data_path, get_tests_output_path -from TTS.api import CS_API, TTS - -OUTPUT_PATH = os.path.join(get_tests_output_path(), "test_python_api.wav") -cloning_test_wav_path = os.path.join(get_tests_data_path(), "ljspeech/wavs/LJ001-0028.wav") - - -is_coqui_available = os.environ.get("COQUI_STUDIO_TOKEN") - - -if is_coqui_available: - - class CS_APITest(unittest.TestCase): - def test_speakers(self): - tts = CS_API() - self.assertGreater(len(tts.speakers), 1) - - def test_emotions(self): - tts = CS_API() - self.assertGreater(len(tts.emotions), 1) - - def test_list_calls(self): - tts = CS_API() - self.assertGreater(len(tts.list_voices()), 1) - self.assertGreater(len(tts.list_speakers()), 1) - self.assertGreater(len(tts.list_all_speakers()), 1) - self.assertGreater(len(tts.list_speakers_as_tts_models()), 1) - - def test_name_to_speaker(self): - tts = CS_API() - speaker_name = tts.list_speakers_as_tts_models()[0].split("/")[2] - speaker = tts.name_to_speaker(speaker_name) - self.assertEqual(speaker.name, speaker_name) - - def test_tts(self): - tts = CS_API() - wav, sr = tts.tts(text="This is a test.", speaker_name=tts.list_speakers()[0].name) - self.assertEqual(sr, 44100) - self.assertGreater(len(wav), 1) - - class TTSTest(unittest.TestCase): - def test_single_speaker_model(self): - tts = TTS(model_name="tts_models/de/thorsten/tacotron2-DDC", progress_bar=False, gpu=False) - - error_raised = False - try: - tts.tts_to_file(text="Ich bin eine Testnachricht.", speaker="Thorsten", language="de") - except ValueError: - error_raised = True - - tts.tts_to_file(text="Ich bin eine Testnachricht.", file_path=OUTPUT_PATH) - - self.assertTrue(error_raised) - self.assertFalse(tts.is_multi_speaker) - self.assertFalse(tts.is_multi_lingual) - self.assertIsNone(tts.speakers) - self.assertIsNone(tts.languages) - - def test_studio_model(self): - tts = TTS(model_name="coqui_studio/en/Zacharie Aimilios/coqui_studio") - tts.tts_to_file(text="This is a test.") - - # check speed > 2.0 raises error - raised_error = False - try: - _ = tts.tts(text="This is a test.", speed=4.0, emotion="Sad") # should raise error with speed > 2.0 - except ValueError: - raised_error = True - self.assertTrue(raised_error) - - # check emotion is invalid - raised_error = False - try: - _ = tts.tts(text="This is a test.", speed=2.0, emotion="No Emo") # should raise error with speed > 2.0 - except ValueError: - raised_error = True - self.assertTrue(raised_error) - - # check valid call - wav = tts.tts(text="This is a test.", speed=2.0, emotion="Sad") - self.assertGreater(len(wav), 0) - - def test_fairseq_model(self): # pylint: disable=no-self-use - tts = TTS(model_name="tts_models/eng/fairseq/vits") - tts.tts_to_file(text="This is a test.") - - def test_multi_speaker_multi_lingual_model(self): - tts = TTS() - tts.load_tts_model_by_name(tts.models[0]) # YourTTS - tts.tts_to_file( - text="Hello world!", speaker=tts.speakers[0], language=tts.languages[0], file_path=OUTPUT_PATH - ) - - self.assertTrue(tts.is_multi_speaker) - self.assertTrue(tts.is_multi_lingual) - self.assertGreater(len(tts.speakers), 1) - self.assertGreater(len(tts.languages), 1) - - def test_voice_cloning(self): # pylint: disable=no-self-use - tts = TTS() - tts.load_tts_model_by_name("tts_models/multilingual/multi-dataset/your_tts") - tts.tts_to_file("Hello world!", speaker_wav=cloning_test_wav_path, language="en", file_path=OUTPUT_PATH) - - def test_voice_conversion(self): # pylint: disable=no-self-use - tts = TTS(model_name="voice_conversion_models/multilingual/vctk/freevc24", progress_bar=False, gpu=False) - tts.voice_conversion_to_file( - source_wav=cloning_test_wav_path, - target_wav=cloning_test_wav_path, - file_path=OUTPUT_PATH, - ) diff --git a/tests/api_tests/test_synthesize_api.py b/tests/api_tests/test_synthesize_api.py deleted file mode 100644 index e7b4f12048..0000000000 --- a/tests/api_tests/test_synthesize_api.py +++ /dev/null @@ -1,25 +0,0 @@ -import os - -from tests import get_tests_output_path, run_cli - - -def test_synthesize(): - """Test synthesize.py with diffent arguments.""" - output_path = os.path.join(get_tests_output_path(), "output.wav") - - # 🐸 Coqui studio model - run_cli( - 'tts --model_name "coqui_studio/en/Torcull Diarmuid/coqui_studio" ' - '--text "This is it" ' - f'--out_path "{output_path}"' - ) - - # 🐸 Coqui studio model with speed arg. - run_cli( - 'tts --model_name "coqui_studio/en/Torcull Diarmuid/coqui_studio" ' - '--text "This is it but slow" --speed 0.1' - f'--out_path "{output_path}"' - ) - - # test pipe_out command - run_cli(f'tts --text "test." --pipe_out --out_path "{output_path}" | aplay')