app.py

from __future__ import annotations

import importlib.util
import os
import time
from pathlib import Path
from typing import Callable, List

import gradio as gr

from libs.asr import predict


def load_inverse_text_function() -> Callable[[str], str]:
    inverse_file = Path("./libs/inverse-text.py")

    if not inverse_file.exists():
        raise FileNotFoundError(f"could not find inverse text file: {inverse_file}")

    spec = importlib.util.spec_from_file_location(
        "inverse_text_module",
        str(inverse_file),
    )

    if spec is None or spec.loader is None:
        raise ImportError(f"could not load inverse text module from: {inverse_file}")

    module = importlib.util.module_from_spec(spec)
    spec.loader.exec_module(module)

    if not hasattr(module, "InverseText"):
        raise AttributeError(f"{inverse_file} must contain function InverseText(text)")

    return module.InverseText


InverseText = load_inverse_text_function()


def load_audio_examples(test_dir: str = "./test") -> List[List[str]]:
    audio_extensions = {
        ".wav",
        ".mp3",
        ".flac",
        ".ogg",
        ".m4a",
        ".aac",
        ".webm",
        ".opus",
    }

    folder = Path(test_dir)

    if not folder.exists():
        return []

    examples = []

    for file_path in sorted(folder.rglob("*")):
        if file_path.is_file() and file_path.suffix.lower() in audio_extensions:
            examples.append([str(file_path)])

    return examples


def transcribe_easier(filepath: str):
    if not filepath:
        return "", "", "0 ms"

    start_time = time.perf_counter()

    result = predict(filepath)

    execute_time_ms = (time.perf_counter() - start_time) * 1000

    if not result.get("success"):
        error = result.get("error", "unknown error")
        return f"ASR error: {error}", "", f"{execute_time_ms:.2f} ms"

    transcribe_text = result.get("transcription", "") or ""

    try:
        inverse_text = InverseText(transcribe_text)
    except Exception as error:
        inverse_text = f"InverseText error: {error}"

    return transcribe_text, inverse_text, f"{execute_time_ms:.2f} ms"


examples = load_audio_examples("./test")


iface = gr.Interface(
    fn=transcribe_easier,
    inputs=gr.Audio(
        sources=["upload", "microphone"],
        type="filepath",
        label="audio",
    ),
    outputs=[
        gr.Textbox(label="transcribe", lines=4),
        gr.Textbox(label="inverse text", lines=4),
        gr.Textbox(label="execute time"),
    ],
    examples=examples,
    cache_examples=False,
    title="Infinity Khmer ASR",
    description="Infinity Khmer ASR demo for Khmer speech recognition develop by @ លោក សយ វិទូ​ was trained with 200 hours",
)


if __name__ == "__main__":
    iface.launch(
        server_name=os.getenv("GRADIO_SERVER_NAME", "0.0.0.0"),
        server_port=int(os.getenv("GRADIO_SERVER_PORT", "7860")),
        share=False,
    )