import sys import os # By using XTTS you agree to CPML license https://coqui.ai/cpml os.environ["COQUI_TOS_AGREED"] = "1" import langid import gradio as gr from TTS.api import TTS HF_TOKEN = os.environ.get("HF_TOKEN") from huggingface_hub import HfApi # will use api to restart space on a unrecoverable error api = HfApi(token=HF_TOKEN) repo_id = "coqui/xtts" tts = TTS("tts_models/multilingual/multi-dataset/xtts_v1") tts.to("cuda") DEVICE_ASSERT_DETECTED=0 DEVICE_ASSERT_PROMPT=None DEVICE_ASSERT_LANG=None def predict(prompt, language, audio_file_pth, mic_file_path, use_mic,no_lang_auto_detect, agree): if agree == True: supported_languages=["en","es","fr","de","it","pt","pl","tr","ru","nl","cs","ar","zh-cn"] if language not in supported_languages: gr.Warning("Language you put in is not in is not in our Supported Languages, please choose from dropdown") return ( None, None, ) language_predicted=langid.classify(prompt)[0].strip() # strip need as there is space at end! if language_predicted == "zh": #we use zh-cn language_predicted = "zh-cn" print(f"Detected language:{language_predicted}, Chosen language:{language}") if len(prompt)>15: #allow any language for short text as some may be common if language_predicted != language and not no_lang_auto_detect: #Please duplicate and remove this check if you really want this #Or auto-detector fails to identify language (which it can on pretty short text or mixed text) gr.Warning(f"It looks like your text isn’t the language you chose , if you’re sure the text is the same language you chose, please check disable language auto-detection checkbox" ) return ( None, None, ) if use_mic == True: if mic_file_path is not None: speaker_wav=mic_file_path else: gr.Warning("Please record your voice with Microphone, or uncheck Use Microphone to use reference audios") return ( None, None, ) else: speaker_wav=audio_file_pth if len(prompt)<2: gr.Warning("Please give a longer prompt text") return ( None, None, ) if len(prompt)>200: gr.Warning("Text length limited to 200 characters for this demo, please try shorter text. You can clone this space and edit code for your own usage") return ( None, None, ) global DEVICE_ASSERT_DETECTED if DEVICE_ASSERT_DETECTED: global DEVICE_ASSERT_PROMPT global DEVICE_ASSERT_LANG #It will likely never come here as we restart space on first unrecoverable error now print(f"Unrecoverable exception caused by language:{DEVICE_ASSERT_LANG} prompt:{DEVICE_ASSERT_PROMPT}") try: tts.tts_to_file( text=prompt, file_path="output.wav", speaker_wav=speaker_wav, language=language, ) except RuntimeError as e : if "device-side assert" in str(e): # cannot do anything on cuda device side error, need tor estart print(f"Exit due to: Unrecoverable exception caused by language:{language} prompt:{prompt}", flush=True) gr.Warning("Unhandled Exception encounter, please retry in a minute") print("Cuda device-assert Runtime encountered need restart") if not DEVICE_ASSERT_DETECTED: DEVICE_ASSERT_DETECTED=1 DEVICE_ASSERT_PROMPT=prompt DEVICE_ASSERT_LANG=language # HF Space specific.. This error is unrecoverable need to restart space api.restart_space(repo_id=repo_id) else: print("RuntimeError: non device-side assert error:", str(e)) raise e return ( gr.make_waveform( audio="output.wav", ), "output.wav", ) else: gr.Warning("Please accept the Terms & Condition!") return ( None, None, ) title = "Coqui🐸 XTTS" description = """ XTTS is a Voice generation model that lets you clone voices into different languages by using just a quick 3-second audio clip.
XTTS is built on previous research, like Tortoise, with additional architectural innovations and training to make cross-language voice cloning and multilingual speech generation possible.
This is the same model that powers our creator application Coqui Studio as well as the Coqui API. In production we apply modifications to make low-latency streaming possible.
Leave a star on the Github 🐸TTS, where our open-source inference and training code lives.

For faster inference without waiting in the queue, you should duplicate this space and upgrade to GPU via the settings.
Duplicate Space

Language Selectors: Arabic: ar, Brazilian Portuguese: pt , Chinese: zh-cn, Czech: cs,
Dutch: nl, English: en, French: fr, Italian: it, Polish: pl,
Russian: ru, Spanish: es, Turkish: tr

""" article = """

By using this demo you agree to the terms of the Coqui Public Model License at https://coqui.ai/cpml

""" examples = [ [ "Once when I was six years old I saw a magnificent picture", "en", "examples/female.wav", None, False, False, True, ], [ "Lorsque j'avais six ans j'ai vu, une fois, une magnifique image", "fr", "examples/male.wav", None, False, False, True, ], [ "Als ich sechs war, sah ich einmal ein wunderbares Bild", "de", "examples/female.wav", None, False, False, True, ], [ "Cuando tenía seis años, vi una vez una imagen magnífica", "es", "examples/male.wav", None, False, False, True, ], [ "Quando eu tinha seis anos eu vi, uma vez, uma imagem magnífica", "pt", "examples/female.wav", None, False, False, True, ], [ "Kiedy miałem sześć lat, zobaczyłem pewnego razu wspaniały obrazek", "pl", "examples/male.wav", None, False, False, True, ], [ "Un tempo lontano, quando avevo sei anni, vidi un magnifico disegno", "it", "examples/female.wav", None, False, False, True, ], [ "Bir zamanlar, altı yaşındayken, muhteşem bir resim gördüm", "tr", "examples/female.wav", None, False, False, True, ], [ "Когда мне было шесть лет, я увидел однажды удивительную картинку", "ru", "examples/female.wav", None, False, False, True, ], [ "Toen ik een jaar of zes was, zag ik op een keer een prachtige plaat", "nl", "examples/male.wav", None, False, False, True, ], [ "Když mi bylo šest let, viděl jsem jednou nádherný obrázek", "cs", "examples/female.wav", None, False, False, True, ], [ "当我还只有六岁的时候， 看到了一副精彩的插画", "zh-cn", "examples/female.wav", None, False, False, True, ], ] gr.Interface( fn=predict, inputs=[ gr.Textbox( label="Text Prompt", info="One or two sentences at a time is better", value="Hi there, I'm your new voice clone. Try your best to upload quality audio", ), gr.Dropdown( label="Language", info="Select an output language for the synthesised speech", choices=[ "en", "es", "fr", "de", "it", "pt", "pl", "tr", "ru", "nl", "cs", "ar", "zh-cn", ], max_choices=1, value="en", ), gr.Audio( label="Reference Audio", info="Click on the ✎ button to upload your own target speaker audio", type="filepath", value="examples/female.wav", ), gr.Audio(source="microphone", type="filepath", info="Use your microphone to record audio", label="Use Microphone for Reference"), gr.Checkbox(label="Check to use Microphone as Reference", value=False, info="Notice: Microphone input may not work properly under traffic",), gr.Checkbox(label="Do not use language auto-detect", value=False, info="Check to disable language detection, if your really intend to",), gr.Checkbox( label="Agree", value=False, info="I agree to the terms of the Coqui Public Model License at https://coqui.ai/cpml", ), ], outputs=[ gr.Video(label="Waveform Visual"), gr.Audio(label="Synthesised Audio"), ], title=title, description=description, article=article, examples=examples, ).queue().launch(debug=True)