Turn an article or PDF into a two-host podcast
To turn an article or a PDF into a podcast, you give the text to a tool that writes a conversation between two hosts, check that script against the source, and voice it with two different AI voices into one audio file. In Speecho that is one button in Text to speech: the hosts talk your text through in its own order, the script opens in an editor where you can fix every line, and a ten-minute episode costs about a dollar in Standard voices. Here is what a two-host episode is good for, how to pick the text, and how to make sure the hosts say only what the text says.
What a two-host episode is good for
- Listening on the move. The long read you saved last week and never opened becomes ten minutes on a commute, a run or the washing-up.
- Studying. A chapter turned into questions and answers makes its structure audible: one host asks what the point is, the other explains it, and you hear the argument instead of a wall of sentences. It works as revision, not as a replacement for reading.
- Newsletters. An audio edition of each issue for the readers who prefer to listen.
- Internal updates. A policy change, a quarterly report or release notes as a short episode the team will actually play.
Why two voices instead of one narrator? A conversation signposts: a question announces what comes next, a rephrasing repeats the key idea, and the change of voice keeps attention. A single voice reading a dense article word for word is tiring after a few minutes. If you do want the text read exactly as written, use plain text to speech or PDF to audio instead.
Picking the text
The best sources have an argument or a story: explainers, essays, reports, a book chapter, your own notes from a course. Tables, code, long lists of figures and legal texts where every word counts make poor conversation.
A few things help before you start:
- Clean the extract. Text from a PDF often carries page numbers, running headers and a list of references. Delete them, and cut any section you do not want covered.
- Mind the size. The text can be 300 to 60,000 characters long, roughly 50 to 10,000 words. Only the first 60,000 characters of a longer text are used, so split a book chapter by chapter.
- Check the rights. More on that at the end.
Make the episode in Speecho
- Add the text. Open Text to speech and paste the text into the box, or choose “Open a file” for a PDF, Word or EPUB document. The file is read in your browser.
- Press “Make a podcast from text”. It appears in Dialogue mode, and next to the Text / Dialogue switch once the text is long enough.
- Set up the hosts. There are two. Speecho picks their voices for you, a woman and a man from the voices made for conversation, and you can change both later. Names are optional; without one, a host goes by the voice’s name.
- Choose the length: about 5, 10 or 20 minutes.
- Choose the language. By default the script is in the language of the text. Pick another to turn, say, a German report into an English episode.
- Press “Write the script”. The button shows the price. Writing takes 10 to 30 seconds.
- Check the script in the Dialogue editor, where it opens with both hosts and every line. See the next section.
- Voice it. The whole dialogue becomes one MP3. A long dialogue is voiced in portions: the first 5,000 characters or so are voiced first, so you can listen before paying for the rest, then “voice the rest” from the card.
Check the script before you voice it
The hosts are instructed to stick to the text, follow its order and discuss it rather than read it aloud. The script is still written by AI, and AI can get details wrong. Five minutes of checking saves a re-recording:
- Facts and numbers. Compare every figure, date and quote with the source. Delete any claim the text does not make.
- Order and emphasis. The script follows the source. If one section matters more to your listeners, give it more lines; if one matters less, delete some.
- Names. Make sure people, companies and places are spelled as in the source. If a voice mispronounces a name in the preview, write it the way it sounds in that line.
- The opening. The first line should name the source correctly: who wrote it and where it was published.
- Tone. Cut lines that only gush (“That’s fascinating!”) and keep the ones that move the topic forward.
Editing is free: nothing is charged until you voice the script.
Choosing the voices
- Make them easy to tell apart. Two voices of different gender or clearly different pitch let listeners follow who is talking without names.
- Native voices for the language. A host with a native accent in the episode’s language sounds natural, so pick voices made for that language.
- Pick a tier. Standard is fine for internal updates and personal listening. Enhanced sounds more natural and is the usual choice for anything you publish. Premium is the most expressive.
- Preview before voicing. Listen to each voice’s sample in the editor first.
How long should the episode be?
| Length | Good for | Script size |
|---|---|---|
| About 5 minutes | One article or newsletter issue | About 750 words |
| About 10 minutes | A long read, a report, release notes | About 1,500 words, around 9,000 characters |
| About 20 minutes | A book chapter, a long report | About 3,000 words |
The episode discusses the text instead of reading it, so its length does not depend on the source: a 6,000-word chapter can become a 20-minute conversation that covers the main points, not 45 minutes of narration. The length is a target, not a promise; a 20-minute setting can come out a few minutes shorter.
Subtitles for a video version
Many podcasts also go out as video, often a still image or a waveform with captions. Clips in Enhanced and Premium voices come with subtitle files: pick “Subtitles (SRT)” or VTT from the arrow next to the MP3 download, or “Subtitles with names (SRT)” for a dialogue where the speaker’s name should appear. The timing follows the audio, so the file drops straight into a video editor or a YouTube upload. Standard voices have no subtitle files.
Fixing one line afterwards
Heard a wrong word after voicing? You do not have to redo the episode. Each line of a voiced dialogue has “Voice this line again”: pick another voice, fix a word, or add a direction like [whispers]. The rest stays as it is, and the price of that one line is shown before you confirm.
What it costs
Two things are paid separately, and both prices are shown before anything is charged:
- The script: 15 credits for a script of up to 5 minutes, 25 for up to 10, 40 for up to 20, plus 1 credit per 3,000 characters of the source text. A short text is not stretched to the minutes you pick and is charged for the script it really gets. A 6,000-character article costs 17 credits at 5 minutes or 27 at 10; a 30,000-character chapter at 10 minutes costs 35.
- The voicing: per 1,000 characters of script, 50 credits in Standard, 110 in Enhanced, 160 in Premium.
A 10-minute episode, about 9,000 characters of script, comes to roughly 450 credits in Standard and 990 in Enhanced. Credits come in packs: $5 for 2,400, $10 for 6,000, $25 for 16,800 and $50 for 36,000. So a 10-minute episode costs about $0.94 in Standard voices and $2 in Enhanced on the smallest pack, and less on the bigger ones. There is no subscription, and credits never expire. See pricing for the details.
Where you can publish it
Most platforms accept AI-voiced podcasts, some with a condition. Apple Podcasts, for example, asks for the use of synthetic voices to be disclosed in the episode itself and in its metadata. A sentence at the start (“This episode is voiced by AI hosts and based on an article by…”) and a line in the show notes cover it. The rules for each platform are collected in where AI voices are allowed. Speecho’s voices are licensed stock voices, not copies of real people, and you can use the audio commercially.
Only use texts you have the right to use
A podcast built from someone else’s article is an adaptation of their work. Use your own writing, your company’s documents, texts under a licence that allows adaptations, or works in the public domain. Publishing an episode made from another author’s article or book without their permission is a copyright problem, even with a credit line, and the AI hosts do not change that.
FAQ
Can I make a podcast from a PDF?
Yes. Open the PDF in Text to speech, then press “Make a podcast from text”. The file is read in your browser; a scanned PDF without a text layer goes through the Scan tab first.
Do the AI hosts make things up?
They are instructed to stick to the text and its order, but the script is written by AI and can get details wrong. That is why it opens in the editor first: check facts, numbers and names before voicing.
How long can the source text be?
From 300 to 60,000 characters. A longer text is cut to its first 60,000 characters, so split books by chapter.
How much does a 10-minute episode cost?
About 450 credits in Standard voices or 990 in Enhanced, plus 15 to 40 credits and 1 per 3,000 characters for the script. On the $5 pack, from a 6,000-character article (a 27-credit script), that is about $0.99 or $2.12.
Can I change one line after the episode is voiced?
Yes. Use “Voice this line again” on that line to change the voice, a word or the delivery. The rest of the episode stays as it is.
