如何使用 ElevenLabs 建立 AI Podcast 生成器

Why an AI‑Powered Podcast?

Podcasts have exploded in popularity, but producing a full episode still means spending hours on research, script‑writing, recording, and editing. What if you could automate the whole pipeline? By combining a large‑language model (LLM) for content generation with a state‑of‑the‑art text‑to‑speech (TTS) service, you can spin up a fully voiced episode in minutes.

In this guide we’ll build a simple AI Podcast Generator that:

1. Takes a topic prompt, asks an LLM (e.g., OpenAI’s gpt‑4o) to write a short script.
2. Sends each paragraph to ElevenLabs – a high‑quality TTS platform that supports voice cloning.
3. Concatenates the audio files into a single MP3 ready for publishing.

Everything is done with plain Python and a few command‑line tools, so you can drop it into a CI job, a serverless function, or your local dev environment.

Prerequisites

If you don’t have an ElevenLabs account yet, grab a free trial through this affiliate link: https://try.elevenlabs.io/kr07zfuqn1bp. It gives you 10 k characters of TTS credit and access to a handful of premium voices.

1. Get Your API Keys

# Save your keys as environment variables (never hard‑code them!)
export OPENAI_API_KEY="sk-..."
export ELEVENLABS_API_KEY="your-elevenlabs-key"

Both services use bearer tokens in the Authorization header, so keeping them in the environment is the safest approach.

2. Generate a Podcast Script

Below is a minimal wrapper around the OpenAI chat completion endpoint. It accepts a topic and returns a concise script split into paragraphs.

import os
import json
import httpx

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
HEADERS = {"Authorization": f"Bearer {OPENAI_API_KEY}"}

def generate_script(topic: str, paragraphs: int = 5) -> list[str]:
    prompt = f"""You are a podcast host. Write a {paragraphs}-paragraph script about **{topic}**.
Each paragraph should be a self‑contained speaking segment (≈40‑60 words). Use a conversational tone, include a brief intro and a wrap‑up."""
    payload = {
        "model": "gpt-4o",
        "messages": [{"role": "user", "content": prompt}],
        "temperature": 0.7,
    }
    resp = httpx.post("https://api.openai.com/v1/chat/completions", headers=HEADERS, json=payload)
    resp.raise_for_status()
    content = resp.json()["choices"][0]["message"]["content"]
    # Split on double newlines to get separate paragraphs
    return [p.strip() for p in content.split("\n\n") if p.strip()]

Tip: Adjust temperature for more creative vs. deterministic output. For a news‑style podcast you might set it lower (0.3).

3. Turn Text into Speech with ElevenLabs

ElevenLabs provides a clean REST API that accepts plain text and returns an MP3 buffer. The following function wraps that call.

import pathlib
import uuid

ELEVENLABS_API_KEY = os.getenv("ELEVENLABS_API_KEY")
ELEVEN_HEADERS = {
    "xi-api-key": ELEVENLABS_API_KEY,
    "Content-Type": "application/json",
}

VOICE_ID = "21m00Tcm4TlvDq8ikWAM"  # Example: "Rachel" voice – change to your favorite

def synthesize_paragraph(text: str, output_dir: pathlib.Path) -> pathlib.Path:
    payload = {
        "text": text,
        "voice_settings": {
            "stability": 0.75,
            "similarity_boost": 0.85
        }
    }
    url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}"
    resp = httpx.post(url, headers=ELEVEN_HEADERS, json=payload)
    resp.raise_for_status()
    # Save MP3 with a unique name
    out_path = output_dir / f"{uuid.uuid4()}.mp3"
    out_path.write_bytes(resp.content)
    return out_path

You can experiment with different voice IDs or even upload a custom cloned voice through ElevenLabs’ dashboard. For a truly branded podcast, clone the voice of your host and reference it here – the same API call works with the cloned voice’s ID.

4. Stitch the Segments Together

ffmpeg makes concatenation painless. First, write a temporary file list, then run the command.

import subprocess

def combine_audio(files: list[pathlib.Path], output: pathlib.Path):
    list_path = output.parent / "files.txt"
    with open(list_path, "w", encoding="utf-8") as f:
        for p in files:
            f.write(f"file '{p.as_posix()}'\n")
    # -safe 0 allows absolute paths
    cmd = [
        "ffmpeg", "-y", "-f", "concat", "-safe", "0",
        "-i", str(list_path), "-c", "copy", str(output)
    ]
    subprocess.run(cmd, check=True)
    list_path.unlink()  # clean up

5. The Full Pipeline

Putting everything together:

import pathlib

def create_podcast(topic: str, out_file: pathlib.Path):
    work_dir = pathlib.Path("temp_audio")
    work_dir.mkdir(exist_ok=True)

    # 1️⃣ Generate script
    paragraphs = generate_script(topic)

    # 2️⃣ Synthesize each paragraph
    audio_files = [
        synthesize_paragraph(p, work_dir) for p in paragraphs
    ]

    # 3️⃣ Combine into one episode
    combine_audio(audio_files, out_file)

    # Optional clean‑up
    for f in work_dir.iterdir():
        f.unlink()
    work_dir.rmdir()

if __name__ == "__main__":
    import sys
    if len(sys.argv) != 3:
        print("Usage: python podcast.py <topic> <output.mp3>")
        sys.exit(1)
    create_podcast(sys.argv[1], pathlib.Path(sys.argv[2]))
    print("✅ Podcast generated!")

Run it like:

python podcast.py "The future of quantum computing" episode1.mp3

You’ll end up with an MP3 that sounds like a professional host, ready to upload to Spotify, Apple Podcasts, or any platform you prefer.

6. Going Further

All of these extensions still rely on the same core building blocks: an LLM for content, ElevenLabs for realistic TTS, and a tiny amount of glue code.

7. Why ElevenLabs Rocks for This Use‑Case

• Human‑like prosody – The generated speech includes natural pauses, emphasis, and intonation that older TTS engines lack.
• Voice cloning – Upload a few recordings and create a unique “host” voice that matches your brand.
• Scalable pricing – Pay‑as‑you‑go credits make it cheap to spin up dozens of episodes per month.

If you’re looking for a plug‑and‑play solution that delivers studio‑grade audio without a massive learning curve, ElevenLabs is the go‑to service. Grab your free credits here: https://try.elevenlabs.io/kr07zfuqn1bp.

Wrap‑Up

You now have a fully functional AI Podcast Generator that can:

1. Write a script from a simple topic prompt.
2. Render each paragraph with lifelike speech.
3. Stitch everything together into a publishable episode.

The code is deliberately straightforward so you can adapt it to your own workflow—whether you want to add background music, integrate a transcription service, or schedule daily releases.

Ready to give your listeners a voice they’ll love? Try ElevenLabs today, spin up a few episodes, and watch how quickly you can go from idea to on‑air content. Happy coding!