GUIDE

Article to audio: listen versions for blogs and news sites

Many readers would rather listen: on the way to work, while cooking, or simply because it is easier on the eyes. A listen version of each article meets them there, but recording every post is slow, and cloud speech services charge for every character.

SuperClonica gives you natural voices and a free engine that turns articles into MP3 files on your own computer or server. This guide shows how to prepare the text, how to automate the audio for every new post and which voice to choose.

Browse voices Install the SuperClonica engine

What you need

Generate audio once per version of the article and store the file with the post; regenerate it only when the text changes.

  • The article as plain text: no HTML, no Markdown symbols.
  • The free SuperClonica engine on any computer with Python 3.10 or newer, or its Docker image on your server.
  • A neutral voice: one of the ten stock voices of the model or your own trained voice.
  • No graphics card: an ordinary processor is enough.

Hear a paragraph free online

Before you automate anything, copy a paragraph of a real article and paste it on the page of any voice, 30 to 100 characters, choose the language and press Voice it. In a few seconds you hear how your posts will sound, with no account.

Prepare the article text

The cleaner the text, the less anyone needs to listen to every file before it goes live.

  • Strip HTML and Markdown: links, image captions and code would otherwise be read aloud.
  • Turn headings into short sentences and end each with a full stop, so the voice pauses.
  • Write abbreviations, product names and unusual numbers the way they should be read.
  • Leave out tables, long URLs and embedded widgets, or describe them in one sentence.

One command, or an API on your server

In a build step or a publishing script, run the engine on the prepared text: superclonica --text-file post.txt --voice M3 --lang en --normalize --output post.mp3. It writes an MP3 straight away, evens the loudness to −18 LUFS and joins long articles into one file. If your site is not written in Python, run the engine as a local HTTP service: docker run -d -p 8000:8000 -v superclonica-models:/models superclonica/superclonica:cpu, then send POST /synthesize with the text, the language and the voice, and store the WAV it returns or convert it to MP3. Either way the text never leaves your own infrastructure and there is no fee per character.

Choose the voice and be open about it

For a publication, choose a neutral, clear voice: the ten stock voices of the model, M1 to M5 and F1 to F5, come with the free engine. For a recognizable house voice, train one from a clean recording of a host who agreed, on a computer with an NVIDIA graphics card, or order it on the exchange. Do not use character or parody voices for news. Tell listeners that the audio is generated, and keep the written article accessible: a listen version is an addition, not a replacement.