app.py

import torch
import gradio as gr
import ffmpeg
import numpy as np
import whisper

MODEL_NAME = "large-v3"
SAMPLE_RATE = 16000

device = "cuda" if torch.cuda.is_available() else "cpu"
model = whisper.load_model(MODEL_NAME).to(device)

def load_audio(file):
    try:
        # This launches a subprocess to decode audio while down-mixing and resampling as necessary.
        # Requires the ffmpeg CLI and `ffmpeg-python` package to be installed.
        out, _ = (
            ffmpeg.input(file, threads=0)
            .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=SAMPLE_RATE)
            .run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
        )
    except ffmpeg.Error as e:
        raise RuntimeError(f"Failed to load audio: {e.stderr.decode()}") from e

    return np.frombuffer(out, np.int16).flatten().astype(np.float32) / 32768.0

def transcribe(audio_file, task):
    if audio_file is None:
        raise gr.Error("No audio file submitted! Please upload an audio file before submitting your request.")

    try:
        # Load audio
        audio = load_audio(audio_file.name)

        # Transcribe
        result = model.transcribe(audio, task=task, language="en")

        # Format output
        output = ""
        for segment in result["segments"]:
            start_time = segment["start"]
            end_time = segment["end"]
            text = segment["text"]
            output += f"[{format_timestamp(start_time)} -> {format_timestamp(end_time)}] {text}\n"

        return output

    except Exception as e:
        raise gr.Error(f"Error processing audio file: {str(e)}")

def format_timestamp(seconds):
    minutes, seconds = divmod(seconds, 60)
    hours, minutes = divmod(minutes, 60)
    return f"{int(hours):02d}:{int(minutes):02d}:{seconds:.2f}"

# Use specific Gradio components
audio_input = gr.components.File(label="Audio file", file_types=["audio"])
task_input = gr.components.Radio(["transcribe", "translate"], label="Task", default="transcribe")
output = gr.components.Textbox(label="Transcription with Timestamps")

demo = gr.Interface(
    fn=transcribe,
    inputs=[audio_input, task_input],
    outputs=output,
    title=f"Whisper {MODEL_NAME}: Transcribe Audio with Timestamps",
    description=(
        f"Transcribe audio files with Whisper {MODEL_NAME}. "
        "Upload an audio file and choose whether to transcribe or translate. "
        "The output includes timestamps for each transcribed segment."
    ),
)

if __name__ == "__main__":
    demo.launch()