Spaces:
Running
Running
| import torch | |
| torch.manual_seed(160923) | |
| import gradio as gr | |
| import torch.cuda | |
| from huggingface_hub import hf_hub_download | |
| from InferenceInterfaces.ControllableInterface import ControllableInterface | |
| from Utility.utils import float2pcm | |
| from Utility.utils import load_json_from_path | |
| class TTSWebUI: | |
| def __init__(self, | |
| gpu_id="cpu", | |
| title="Phoneme Synthesis with Neutral Accent and Many Speakers", | |
| article="Put in a string of IPA characters and have it pronounced in a way that is averaged across many languages. Use ~ to get a pause and include any punctuation marks you would normally use. If you enable the checkbox, the model will take much longer, but the result will be spoken by 10 artificial voices at the same time. <br>", | |
| tts_model_path=None, | |
| vocoder_model_path=None, | |
| embedding_gan_path=None, | |
| available_artificial_voices=10 # be careful with this, if you want too many, it might lead to an endless loop | |
| ): | |
| path_to_iso_list = hf_hub_download(repo_id="Flux9665/ToucanTTS", filename="iso_to_fullname.json") | |
| iso_to_name = load_json_from_path(path_to_iso_list) | |
| text_selection = [f"{iso_to_name[iso_code]} ({iso_code})" for iso_code in iso_to_name] | |
| # accent_selection = [f"{iso_to_name[iso_code]} Accent ({iso_code})" for iso_code in iso_to_name] | |
| self.controllable_ui = ControllableInterface(gpu_id=gpu_id, | |
| available_artificial_voices=available_artificial_voices, | |
| tts_model_path=tts_model_path, | |
| vocoder_model_path=vocoder_model_path, | |
| embedding_gan_path=embedding_gan_path) | |
| self.iface = gr.Interface(fn=self.read, | |
| inputs=[gr.Textbox(lines=2, | |
| placeholder="put in IPA symbols here...", | |
| value="~tə ɡɛt ɐ pˈɔːz~ plˈeɪs ɐ tˈɪldə sˈɪmbəl. jˈuːs pʌŋktʃuːˈeɪʃən~ æz ɪf ðɪs wʌz tˈɛkst.~#", | |
| label="IPA Input"), | |
| gr.Slider(minimum=0.0, maximum=0.8, step=0.1, value=0.5, label="Prosody Creativity"), | |
| gr.Checkbox(value=False, label="Speak in many Voices"), | |
| gr.Audio(type="filepath", show_label=True, container=True, label="[OPTIONAL] Voice to Clone (if left empty, will use an artificial voice instead)"), | |
| # gr.Slider(minimum=0.5, maximum=1.5, step=0.1, value=1.0, label="Pitch Variance Scale"), | |
| # gr.Slider(minimum=0.5, maximum=1.5, step=0.1, value=1.0, label="Energy Variance Scale"), | |
| # gr.Slider(minimum=-10.0, maximum=10.0, step=0.1, value=0.0, label="Voice Depth") | |
| ], | |
| outputs=[gr.Audio(type="numpy", label="Speech")], | |
| title=title, | |
| allow_flagging="never", | |
| description=article, | |
| theme=gr.themes.Ocean(primary_hue="amber", secondary_hue="orange")) | |
| self.iface.launch() | |
| def read(self, | |
| prompt, | |
| prosody_creativity, | |
| voice_seed, | |
| reference_audio, | |
| ): | |
| sr, wav = self.controllable_ui.read("~" + prompt.replace(",", "~") + "~#", | |
| reference_audio, | |
| voice_seed, | |
| prosody_creativity, | |
| -24.) | |
| return (sr, float2pcm(wav)) | |
| if __name__ == '__main__': | |
| TTSWebUI(gpu_id="cuda" if torch.cuda.is_available() else "cpu") | |