import sys import os,stat import subprocess import random from zipfile import ZipFile import uuid # By using XTTS you agree to CPML license https://coqui.ai/cpml os.environ["COQUI_TOS_AGREED"] = "1" # langid is used to detect language for longer text # Most users expect text to be their own language, there is checkbox to disable it import langid import gradio as gr from TTS.api import TTS from TTS.tts.configs.xtts_config import XttsConfig from TTS.tts.models.xtts import Xtts from TTS.utils.generic_utils import get_user_data_dir 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" # Use never ffmpeg binary for Ubuntu20 to use denoising for microphone input print("Export newer ffmpeg binary for denoise filter") ZipFile("ffmpeg.zip").extractall() print("Make ffmpeg binary executable") st = os.stat('ffmpeg') os.chmod('ffmpeg', st.st_mode | stat.S_IEXEC) # Load TTS from TTS.utils.manage import ModelManager import torch model_name = "tts_models/multilingual/multi-dataset/xtts_v2" ModelManager().download_model(model_name) model_path = os.path.join(get_user_data_dir("tts"), model_name.replace("/", "--")) print("XTTS downloaded") tts = TTS(model_name) if torch.cuda.is_available(): tts.to("cuda") else: tts.to("cpu") # This is for debugging purposes only DEVICE_ASSERT_DETECTED=0 DEVICE_ASSERT_PROMPT=None DEVICE_ASSERT_LANG=None def predict(prompt, language, audio_file_pth, mic_file_path, use_mic, voice_cleanup, no_lang_auto_detect, agree,): if agree == True: supported_languages=["en","es","fr","de","it","pt","pl","tr","ru","nl","cs","ar","zh-cn","ja","ko","hu"] 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, None, ) language_predicted=langid.classify(prompt)[0].strip() # strip need as there is space at end! # tts expects chinese as zh-cn if language_predicted == "zh": #we use zh-cn language_predicted = "zh-cn" print(f"Detected language:{language_predicted}, Chosen language:{language}") # After text character length 15 trigger language detection if len(prompt)>15: # allow any language for short text as some may be common # If user unchecks language autodetection it will not trigger # You may remove this completely for own use 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, 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, None, ) else: speaker_wav=audio_file_pth # Filtering for microphone input, as it has BG noise, maybe silence in beginning and end # This is fast filtering not perfect # Apply all on demand lowpassfilter=denoise=trim=loudness=True if lowpassfilter: lowpass_highpass="lowpass=8000,highpass=75," else: lowpass_highpass="" if trim: # better to remove silence in beginning and end for microphone trim_silence="areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02," else: trim_silence="" if (voice_cleanup): try: out_filename = speaker_wav + str(uuid.uuid4()) + ".wav" #ffmpeg to know output format #we will use newer ffmpeg as that has afftn denoise filter shell_command = f"./ffmpeg -y -i {speaker_wav} -af {lowpass_highpass}{trim_silence} {out_filename}".split(" ") command_result = subprocess.run([item for item in shell_command], capture_output=False,text=True, check=True) speaker_wav=out_filename print("Filtered microphone input") except subprocess.CalledProcessError: # There was an error - command exited with non-zero code print("Error: failed filtering, use original microphone input") else: speaker_wav=speaker_wav if len(prompt)<2: gr.Warning("Please give a longer prompt text") return ( None, 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, 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", language=language, speaker_wav=speaker_wav, ) 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", speaker_wav, ) else: gr.Warning("Please accept the Terms & Condition!") return ( None, None, None, ) title = "🐸 XTTSv2 - 3秒语音合成，支持中英双语，告别电音！" description = f""" ##
🌟 - 只需上传3~10秒语音，支持13种语言，中文能力极大增强！
###
🤗 - 使用[Colab笔记本](https://github.com/KevinWang676/Bark-Voice-Cloning)运行；Powered by [Coqui AI](https://coqui.ai/)
###
🌊 - 更多精彩应用，尽在[滔滔AI](http://www.talktalkai.com)；滔滔AI，为爱滔滔！💕
###
😺️☘️ - 猫尾草修改版 - coqui xTTS v2
###
女声示例 - The booms were tearing at the blocks, the rudder was banging to and fro, and the whole ship creaking, groaning, and jumping like a manufactory.
###
男声示例 - It is a pretty little spot there: a grass plateau, running along by the water's edge, and overhung by willows.
""".strip() article = """

注意❗：请不要生成会对个人以及组织造成侵害的内容，此程序仅供科研、学习及个人娱乐使用。

🌊🏞️🎶 - 江水东流急，滔滔无尽声。 明·顾璘

""" examples = [ [ "Once when I was six years old I saw a magnificent picture", "en", "examples/female.wav", None, False, False, False, True, ], [ "Lorsque j'avais six ans j'ai vu, une fois, une magnifique image", "fr", "examples/male.wav", None, False, False, False, True, False, ], [ "Als ich sechs war, sah ich einmal ein wunderbares Bild", "de", "examples/female.wav", None, False, False, False, True, ], [ "Cuando tenía seis años, vi una vez una imagen magnífica", "es", "examples/male.wav", None, False, False, False, True, ], [ "Quando eu tinha seis anos eu vi, uma vez, uma imagem magnífica", "pt", "examples/female.wav", None, False, False, False, True, ], [ "Kiedy miałem sześć lat, zobaczyłem pewnego razu wspaniały obrazek", "pl", "examples/male.wav", None, False, False, False, True, ], [ "Un tempo lontano, quando avevo sei anni, vidi un magnifico disegno", "it", "examples/female.wav", None, False, False, False, True, ], [ "Bir zamanlar, altı yaşındayken, muhteşem bir resim gördüm", "tr", "examples/female.wav", None, False, False, False, True, ], [ "Когда мне было шесть лет, я увидел однажды удивительную картинку", "ru", "examples/female.wav", None, False, 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, False, True, ], [ "Když mi bylo šest let, viděl jsem jednou nádherný obrázek", "cs", "examples/female.wav", None, False, 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, False, True, ], [ "かつて 六歳のとき、素晴らしい絵を見ました", "ja", "examples/female.wav", None, False, True, False, True, ], [ "한번은 내가 여섯 살이었을 때 멋진 그림을 보았습니다.", "ko", "examples/female.wav", None, False, True, False, True, ], [ "Egyszer hat éves koromban láttam egy csodálatos képet", "hu", "examples/male.wav", None, False, True, 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/xiaoxiao(edgetts).mp3", None, False, 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/jenny(edgetts).mp3", None, False, 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/xiaoni(edgetts).mp3", None, False, 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/hsiaochen(edgetts).mp3", None, False, False, False, True, ], ] gr.Interface( fn=predict, inputs=[ gr.Textbox( label="想要合成的文本内容", lines=3, placeholder="想说却还没说的 还很多" ), gr.Dropdown( label="请选择文本内容对应的语言", choices=[ "en", "es", "fr", "de", "it", "pt", "pl", "tr", "ru", "nl", "cs", "ar", "zh-cn", "ja", "ko", "hu" ], max_choices=1, value="zh-cn", ), gr.Audio( label="通过文件上传语音", type="filepath", value="examples/female.wav", ), gr.Audio(source="microphone", type="filepath", label="使用麦克风上传语音", info="移动端更稳定，电脑端可能无法上传", streaming=True, ), gr.Checkbox(label="是否使用麦克风上传语音", value=False, info="默认为否",), gr.Checkbox(label="是否需要去除背景音", value=False, info="默认为否", ), gr.Checkbox(label="不使用自动探测语言", value=False, info="勾选此选项则不使用自动探测语言",), gr.Checkbox( label="使用条款", value=True, info="我承诺：不会利用此程序生成对个人或组织造成侵害的任何内容", ), ], outputs=[ gr.Video(label="为您合成的专属音频"), gr.Audio(label="Synthesised Audio", visible=False), gr.Audio(label="Reference Audio Used", visible=False), ], title=title, description=description, article=article, examples=examples, ).queue().launch(debug=True,show_api=False,server_name="0.0.0.0")