Deepgram Nova-2 vs
OpenAI Whisper API

import { createClient, LiveTranscriptionEvents } from "@deepgram/sdk";
import { createReadStream } from "fs";

const deepgram = createClient(process.env.DEEPGRAM_API_KEY);

async function streamAudioToDeepgram(audioStream) {
  // Establish full-duplex WebSocket connection to Nova-2
  const connection = deepgram.listen.live({
    model: "nova-2",
    punctuate: true,
    smart_format: true,
    diarize: true,
    encoding: "linear16",
    sample_rate: 16000,
    interim_results: true,
    endpointing: 250 // Returns UtteranceEnd event within 250ms of silence
  });

  connection.on(LiveTranscriptionEvents.Open, () => {
    console.log("Connected to Deepgram Nova-2 WebSocket.");
    audioStream.on("data", (chunk) => connection.send(chunk));
  });

  connection.on(LiveTranscriptionEvents.Transcript, (data) => {
    const transcript = data.channel.alternatives[0].transcript;
    if (transcript.length > 0) {
      console.log(`[${data.is_final ? "FINAL" : "PARTIAL"}]: ${transcript}`);
    }
  });

  connection.on(LiveTranscriptionEvents.UtteranceEnd, () => {
    console.log("User finished speaking. Triggering LLM reasoning immediately!");
  });
}

from openai import OpenAI
import os

client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))

def transcribe_whisper_batch(file_path: str):
    # Note: Strict 25MB file limitation applies
    with open(file_path, "rb") as audio_file:
        transcription = client.audio.transcriptions.create(
            model="whisper-1",
            file=audio_file,
            response_format="verbose_json",
            timestamp_granularities=["word"]
        )
    # Billed strictly at $0.0060 per recorded minute
    return transcription.text