app.py

import os, tempfile
import torch
import gradio as gr
from transformers import pipeline
from transformers.pipelines.audio_utils import ffmpeg_read

# -----------------------------
# CONFIG
# -----------------------------
ASR_MODEL = "openai/whisper-large-v3"
BATCH_SIZE = 8
HAS_CUDA = torch.cuda.is_available()
DEVICE = 0 if HAS_CUDA else "cpu"
DTYPE = torch.float16 if HAS_CUDA else torch.float32

asr_pipe = pipeline(
    task="automatic-speech-recognition",
    model=ASR_MODEL,
    device=DEVICE,
    torch_dtype=DTYPE,
    chunk_length_s=30,
)

def _save_text_file(text: str, suffix: str = ".txt") -> str:
    fd, path = tempfile.mkstemp(suffix=suffix)
    with os.fdopen(fd, "w", encoding="utf-8") as f:
        f.write(text)
    return path

def _resolve_path(x):
    """
    รับค่า input ได้ทั้ง:
    - str (filepath)
    - dict ที่มี key 'name' หรือ 'path'
    - gradio FileData (มี .path)
    คืนค่าเป็น filepath เสมอ
    """
    if x is None:
        return None
    if isinstance(x, str):
        return x
    # Gradio v4/v5 บางทีให้เป็น dict
    if isinstance(x, dict):
        return x.get("path") or x.get("name")
    # Gradio FileData
    path = getattr(x, "path", None)
    if path:
        return path
    # เผื่อกรณี edge
    return str(x)

def _transcribe_from_any(file_like, task: str):
    path = _resolve_path(file_like)
    if not path or not os.path.exists(path):
        raise gr.Error("ไม่พบไฟล์ที่อัปโหลด (path ว่างหรือไฟล์หาย)")

    # อ่านเป็น bytes แล้วให้ ffmpeg แปลงเป็นโมโน float32
    with open(path, "rb") as f:
        payload = f.read()
    audio = ffmpeg_read(payload, asr_pipe.feature_extractor.sampling_rate)
    inputs = {"array": audio, "sampling_rate": asr_pipe.feature_extractor.sampling_rate}

    out = asr_pipe(
        inputs,
        batch_size=BATCH_SIZE,
        generate_kwargs={"task": task},  # 'transcribe' = ตามภาษาเดิม, 'translate' = แปลอังกฤษ
        return_timestamps=True,
    )
    text = out["text"]
    return text, _save_text_file(text, ".txt")

def transcribe_mic(mic_path: str, task: str):
    return _transcribe_from_any(mic_path, task)

def transcribe_audio(file_path: str, task: str):
    return _transcribe_from_any(file_path, task)

def transcribe_video(video_file, task: str):
    # รับเป็นไฟล์วิดีโอ (mp4, mov, webm ก็ได้) แล้วให้ ffmpeg_read ดึงเสียง
    return _transcribe_from_any(video_file, task)

# -----------------------------
# UI
# -----------------------------
with gr.Blocks(title="Whisper V3 – Transcriber (Audio + MP4)") as demo:
    gr.Markdown("## 🎙️ Whisper V3 – Record/Upload Audio or MP4 → Transcript → Download (.txt)")

    with gr.Tab("🎤 Microphone"):
        mic_audio = gr.Audio(sources="microphone", type="filepath", label="Record")
        mic_task  = gr.Radio(["transcribe", "translate"], value="transcribe", label="Task")
        mic_text  = gr.Textbox(label="Transcript", lines=10)
        mic_file  = gr.File(label="Download Transcript (.txt)")
        gr.Button("Run").click(transcribe_mic, inputs=[mic_audio, mic_task], outputs=[mic_text, mic_file])

    with gr.Tab("📁 Audio file"):
        up_audio  = gr.Audio(sources="upload", type="filepath", label="Upload audio")
        up_task   = gr.Radio(["transcribe", "translate"], value="transcribe", label="Task")
        up_text   = gr.Textbox(label="Transcript", lines=10)
        up_file   = gr.File(label="Download Transcript (.txt)")
        gr.Button("Run").click(transcribe_audio, inputs=[up_audio, up_task], outputs=[up_text, up_file])

    with gr.Tab("🎬 Video MP4"):
        # ใช้ gr.File เพื่อให้ได้ path ที่นิ่งที่สุด
        up_video  = gr.File(file_count="single", file_types=["video"], label="Upload MP4 / Video")
        vd_task   = gr.Radio(["transcribe", "translate"], value="transcribe", label="Task")
        vd_text   = gr.Textbox(label="Transcript", lines=10)
        vd_file   = gr.File(label="Download Transcript (.txt)")
        gr.Button("Run").click(transcribe_video, inputs=[up_video, vd_task], outputs=[vd_text, vd_file])

demo.queue().launch()