{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Qwen3-ASR Amharic Curated Demo\n",
        "\n",
        "Run each cell in order. The final cell creates a temporary public Gradio link. Keep the notebook running while using the demo."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "%pip install -q -U qwen-asr gradio soundfile"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import os\n",
        "import torch\n",
        "from qwen_asr import Qwen3ASRModel\n",
        "\n",
        "MODEL_ID = \"b1n1yam/qwen3-asr-0.6b-amharic-gold-silver\"\n",
        "DEVICE = \"cuda:0\" if torch.cuda.is_available() else \"cpu\"\n",
        "DTYPE = torch.bfloat16 if torch.cuda.is_available() else torch.float32\n",
        "\n",
        "print(f\"Loading {MODEL_ID} on {DEVICE}\")\n",
        "model = Qwen3ASRModel.from_pretrained(\n",
        "    MODEL_ID,\n",
        "    dtype=DTYPE,\n",
        "    device_map=DEVICE,\n",
        "    attn_implementation=\"eager\",\n",
        "    max_inference_batch_size=1,\n",
        "    max_new_tokens=512,\n",
        ")\n",
        "print(\"Model ready\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import gradio as gr\n",
        "import soundfile as sf\n",
        "\n",
        "MAX_AUDIO_SECONDS = 60\n",
        "\n",
        "def transcribe(audio_path):\n",
        "    if not audio_path:\n",
        "        raise gr.Error(\"Record or upload an audio clip first.\")\n",
        "    duration = sf.info(audio_path).duration\n",
        "    if duration > MAX_AUDIO_SECONDS:\n",
        "        raise gr.Error(f\"Please use audio shorter than {MAX_AUDIO_SECONDS} seconds.\")\n",
        "    result = model.transcribe(audio=audio_path, language=None)\n",
        "    return result[0].text\n",
        "\n",
        "with gr.Blocks(title=\"Qwen3 ASR Amharic Curated\") as demo:\n",
        "    gr.Markdown(\"# Qwen3 ASR Amharic Curated\")\n",
        "    with gr.Row():\n",
        "        audio = gr.Audio(sources=[\"microphone\", \"upload\"], type=\"filepath\", label=\"Amharic audio\")\n",
        "        transcript = gr.Textbox(label=\"Transcript\", lines=10, buttons=[\"copy\"])\n",
        "    button = gr.Button(\"Transcribe\", variant=\"primary\")\n",
        "    button.click(transcribe, inputs=audio, outputs=transcript, concurrency_limit=1)\n",
        "\n",
        "demo.queue(default_concurrency_limit=1).launch(share=True, debug=True)"
      ]
    }
  ]
}