app.py

import spaces # in windows env, delete related to "spaces"
@spaces.GPU
def gpu():
    pass

import asyncio
import datetime
import logging
import os
import time
import traceback

import edge_tts
import gradio as gr
import librosa
import torch
from huggingface_hub import snapshot_download



logging.getLogger("numba").setLevel(logging.WARNING)
logging.getLogger("markdown_it").setLevel(logging.WARNING)
logging.getLogger("urllib3").setLevel(logging.WARNING)
logging.getLogger("matplotlib").setLevel(logging.WARNING)

limitation = os.getenv("SYSTEM") == "spaces"


# Edge TTS
edge_output_filename = "edge_output.mp3"
tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices())
tts_voices = [f"{v['ShortName']}-{v['Gender']}" for v in tts_voice_list]

# RVC models
model_root = snapshot_download(repo_id="NoCrypt/miku_RVC", token=os.getenv("TOKEN", None))
models = [d for d in os.listdir(model_root) if os.path.isdir(f"{model_root}/{d}")]
models.sort()

initial_md = """
![banner that says mikutts](https://huggingface.co/spaces/NoCrypt/mikuTTS/resolve/main/imgs/banner_mikutts.webp)
"""


from main.inference import run_inference_script

def tts(
    model_name,
    speed,
    tts_text,
    tts_voice,
    f0_up_key,
    f0_method,
    index_rate,
    protect,
    filter_radius=3,
    resample_sr=0,
):
    print("------------------")
    print(datetime.datetime.now())
    print("tts_text:")
    print(tts_text)
    print(f"tts_voice: {tts_voice}, speed: {speed}")
    print(f"Model name: {model_name}")
    print(f"F0: {f0_method}, Key: {f0_up_key}, Index: {index_rate}, Protect: {protect}")
    try:
        if limitation and len(tts_text) > 1000:
            print("Error: Text too long")
            return (
                f"Text characters should be at most 1000 in this huggingface space, but got {len(tts_text)} characters.",
                None,
                None,
            )
        t0 = time.time()
        if speed >= 0:
            speed_str = f"+{speed}%"
        else:
            speed_str = f"{speed}%"
        
        # Fix: Extract just the ShortName from the voice selection
        voice_name = tts_voice.split("-")[0]
        
        asyncio.run(
            edge_tts.Communicate(
                tts_text, voice_name, rate=speed_str
            ).save(edge_output_filename)
        )
        t1 = time.time()
        edge_time = t1 - t0
        audio, sr = librosa.load(edge_output_filename, sr=16000, mono=True)
        duration = len(audio) / sr
        print(f"Audio duration: {duration}s")
        if limitation and duration >= 200:
            print("Error: Audio too long")
            return (
                f"Audio should be less than 200 seconds in this huggingface space, but got {duration}s.",
                edge_output_filename,
                None,
            )

        # Fix: Use edge_output_filename as input_path and fix typo in f0_method
        audio_opt, times, tgt_sr = run_inference_script(
            model_name=model_name,
            input_path=edge_output_filename,
            pitch=f0_up_key,
            f0_method=f0_method,  # Fixed typo from f0_metho
            index_rate=index_rate,
            protect=protect,
            filter_radius=filter_radius,
            resample_sr=resample_sr,
        )
        if tgt_sr != resample_sr >= 16000:
            tgt_sr = resample_sr
        info = f"Success. Time: edge-tts: {edge_time}s, npy: {times[0]}s, f0: {times[1]}s, infer: {times[2]}s"
        print(info)
        return (
            info,
            edge_output_filename,
            (tgt_sr, audio_opt),
        )
    except EOFError:
        info = (
            "It seems that the edge-tts output is not valid. "
            "This may occur when the input text and the speaker do not match. "
            "For example, maybe you entered Japanese (without alphabets) text but chose non-Japanese speaker?"
        )
        print(info)
        return info, None, None
    except:
        info = traceback.format_exc()
        print(info)
        return info, None, None




with gr.Blocks() as app:
    gr.Markdown(initial_md)
    gr.Markdown("# MikuTTS V3")
    gr.Markdown("# Modern - Stylish")    
    with gr.Row():
        with gr.Column():
            model_name = gr.Dropdown(
                label="Model",
                choices=models,
                value=models[0],
            )
            f0_key_up = gr.Number(
                label="Tune",
                value=6,
            )
        with gr.Column():
            f0_method = gr.Radio(
                label="Pitch extraction method (pm: very fast, low quality, rmvpe: a little slow, high quality)",
                choices=["pm", "rmvpe"],  # harvest and crepe is too slow
                value="rmvpe",
                interactive=True,
            )
            index_rate = gr.Slider(
                minimum=0,
                maximum=1,
                label="Index rate",
                value=1,
                interactive=True,
            )
            protect0 = gr.Slider(
                minimum=0,
                maximum=0.5,
                label="Protect",
                value=0.33,
                step=0.01,
                interactive=True,
            )
    with gr.Row():
        with gr.Column():
            tts_voice = gr.Dropdown(
                label="Edge-tts speaker (format: language-Country-Name-Gender), make sure the gender matches the model",
                choices=tts_voices,
                allow_custom_value=False,
                value="ja-JP-NanamiNeural-Female",
            )
            speed = gr.Slider(
                minimum=-100,
                maximum=100,
                label="Speech speed (%)",
                value=0,
                step=10,
                interactive=True,
            )
            tts_text = gr.Textbox(label="Input Text", value="こんにちは、私の名前は初音ミクです!")
        with gr.Column():
            but0 = gr.Button("Convert", variant="primary")
            info_text = gr.Textbox(label="Output info", scale=4)
        with gr.Column():
            with gr.Accordion("Edge Voice", open=False):
                edge_tts_output = gr.Audio(label="Edge Voice", type="filepath")
            tts_output = gr.Audio(label="Result")
        but0.click(
            tts,
            [
                model_name,
                speed,
                tts_text,
                tts_voice,
                f0_key_up,
                f0_method,
                index_rate,
                protect0,
            ],
            [info_text, edge_tts_output, tts_output],
        )
    with gr.Row():
        examples = gr.Examples(
            examples_per_page=100,
            examples=[
                ["こんにちは、私の名前は初音ミクです!", "ja-JP-NanamiNeural-Female", 6],
                ["Hello there. My name is Hatsune Miku!","en-CA-ClaraNeural-Female", 6],
                ["Halo. Nama saya Hatsune Miku!","id-ID-GadisNeural-Female", 4],
                ["Halo. Jenengku Hatsune Miku!","jv-ID-SitiNeural-Female", 10],
            ],
            inputs=[tts_text, tts_voice, f0_key_up],
        )

app.launch(theme="NeoPy=Soft", ssr_mode=False)