Edge Function Limit: Download YouTube audio and call Whisper to Transcribe

[sylas-park]

I imported ytdl to download YouTube audio and call Whisper to Transcribe. Although the code works in a separate test.ts, edge functions cannot read or write files. Thus, I cannot access the downloaded audio file and the error occurred.

Therefore, I tried to send 'audio stream or buffer' using different packages other than ytdl rather than read 'audio file', but I haven't succeeded.

As a result, I am thinking of a building an additional Python server to do this process, which unfortunately causes extra server cost.

Do you guys have any solutions to this problem other than building an additional python server?

Aside from solutions, do you guys know why it's happening in the first place?? Is it really because edge functions cannot read or write files?

import "jsr:@supabase/functions-js/edge-runtime.d.ts";
import OpenAI from "npm:openai@^4.28.0";
import ytdl from "npm:@distube/ytdl-core@^4.15.8";

const MODEL_KEY = Deno.env.get("AI_MODEL_KEY");

const openai = new OpenAI({
  apiKey: MODEL_KEY,
});

async function downloadYouTubeAudio(videoId: string): Promise<Uint8Array> {
  try {
    // Step 1: Fetch the audio stream using ytdl-core
    const audioStream = await ytdl(videoId, {
      quality: "lowest",
      filter: "audioonly",
    });

    // Step 2: Convert Node.js readable stream to Deno-compatible buffer
    const chunks: Uint8Array[] = [];
    for await (const chunk of audioStream) {
      chunks.push(chunk);
    }

    // Combine all chunks into a single Uint8Array
    const audioBuffer = new Uint8Array(
      chunks.reduce((total, chunk) => total + chunk.length, 0),
    );
    chunks.reduce((offset, chunk) => {
      audioBuffer.set(chunk, offset);
      return offset + chunk.length;
    }, 0);

    console.log("Audio downloaded successfully!");
    return audioBuffer;
  } catch (error) {
    console.error("Error:", error);
    throw error;
  }
}

async function transcribeAudio(audioStream: Uint8Array) {
  // Convert the stream to a proper audio file format
  const file = new File([audioStream], "audio.mp3", { type: "audio/mpeg" });

  const transcription = await openai.audio.transcriptions.create({
    file: file,
    model: "whisper-1",
    response_format: "verbose_json",
    timestamp_granularities: ["word"],
  });
  return transcription;
}

Deno.serve(async (req) => {
  // Get the request method
  const { method } = req;

  switch (method) {
    case "POST":
      try {
        const { videoId } = await req.json();
        const audioData = await downloadYouTubeAudio(videoId);
        const transcription = await transcribeAudio(audioData);
        console.log(audioData);
        console.log(transcription);

        return new Response(
          JSON.stringify({ transcription }),
          { headers: { "Content-Type": "application/json" } },
        );
      } catch (error) {
        console.error("Error:", error);
        return new Response(
          JSON.stringify({
            error: error instanceof Error
              ? error.message
              : "Unknown error occurred",
          }),
          { status: 500, headers: { "Content-Type": "application/json" } },
        );
      }

    default:
      return new Response(
        JSON.stringify({ error: "Method not allowed" }),
        {
          status: 405,
          headers: { "Content-Type": "application/json" },
        },
      );
  }
});

[GaryAustin1]

The ability to use the Deno filesystem was just added in Dec for Supabase Edge functions. Make sure you are on latest version of CLI.
https://supabase.com/docs/guides/functions/ephemeral-storage
It does not support all operations and is limited in size.