import spaces import random import gradio as gr from css.utils import * # 预置语音生成 def preset(): sound_choices_values = ['中文女', '中文男', '英文女', '英文男', '日语男', '粤语女', '韩语女'] sound_choices = ['Chinese Female', 'Chinese Male', 'Japanese Male', 'English Female', 'English Male', 'Cantonese Female', 'Korean Female'] sound_choices_dict = {k:v for k, v in zip(sound_choices, sound_choices_values)} def random_seed(): return random.randint(1, 100000000) @spaces.GPU def generate_audio(_sound_radio, _synthetic_input_textbox, _seed): _sound_radio = sound_choices_dict[_sound_radio] print(_sound_radio, _synthetic_input_textbox, _seed) if _synthetic_input_textbox == '': # gr.Warning('合成文本为空,您是否忘记输入合成文本?') gr.Warning('The synthesis text is empty, did you forget to input the synthesis text?') return (target_sr, default_data) set_all_random_seed(_seed) if use_instruct(_synthetic_input_textbox): model = cosyvoice_instruct else: model = cosyvoice_sft output = model.inference_sft(_synthetic_input_textbox, _sound_radio) audio_data = postprocess(output['tts_speech']).numpy().flatten() return (target_sr, audio_data) with gr.Column(): sound_radio = gr.Radio(choices=sound_choices, value=sound_choices[0], label="Select Preset Voice") #label="选择预置音色") with gr.Column(): synthetic_input_textbox = gr.Textbox(label="Input Synthesis Text") gr.Examples( label="Example Text", examples=example_tts_text, inputs=[synthetic_input_textbox]) with gr.Accordion(label="Random Seed"): with gr.Row(): with gr.Column(scale=1, min_width=180): seed_button = gr.Button(value="\U0001F3B2 Shuffle randomly", elem_classes="full-height") with gr.Column(scale=10): seed = gr.Number(show_label=False, value=0, container=False, elem_classes="full-height") with gr.Column(): generate_button = gr.Button("Generate audio", variant="primary", size="lg") with gr.Column(): output_audio = gr.Audio(label="Synthesize Audio") seed_button.click(fn=random_seed, outputs=[seed]) generate_button.click(fn=generate_audio, inputs=[sound_radio, synthetic_input_textbox, seed], outputs=[output_audio])