How to make a two-voice AI podcast from a script (step by step)

Why a two-voice format works

Conversations are easier to follow than a single narrator. A host asks the questions your listeners would ask, and a guest answers them. With AI dialogue you can produce that format from a script, without booking guests or a studio, and in many languages.

Step 1: Write the script

Start every line with the speaker. In VoiceLamp, speaker1: is the host and speaker2: is the guest:

speaker1: Welcome to the show! Really glad you could join us today.
speaker2: Thanks for having me. I've been looking forward to this.
speaker1: Let's dive in. What's changing fastest in your industry?

You can also write Host: and Guest:, or the names you give the speakers. A line without a label continues the previous turn, so a colon inside a sentence is fine.

Step 2: Add emotion

Add performance tags inside a line, such as [laughs], [whispers], or [excited]. Use them sparingly; one or two per minute keeps the delivery natural.

Step 3: Pick two different voices

Choose a voice for the host and the guest and play the previews. Pick voices that are clearly different, such as different pitch or accent, so listeners always know who is speaking. Give each speaker a display name; it is used in the preview and in the downloadable script.

Step 4: Generate and download

Choose the language (84 are supported, or use auto-detect), add an episode title, and check the parsed preview: it shows each turn, the character count, and the estimated length. Select Generate Podcast to see the exact price, then approve. You get one MP3 episode and can download the script as text for your show notes.

Scripts can be up to 5,000 characters, about five to six minutes of audio. Long scripts are recorded in parts and joined into one file automatically.

What it costs

Podcasts are priced by script length, at about 68 credits per character. A 3,000-character episode, roughly three minutes, costs about 204,000 credits. You always see the exact price before generating, and failed jobs are refunded automatically.

Before you publish

• Listen through once and fix names or numbers the voices read oddly by spelling them out.
• Add intro music made in the sound studio.
• Label the episode as AI-generated where your platform or local law requires it.