Converting text to MP3 takes about thirty seconds once you know the steps. This guide walks through it on tts.audio, covers PDF input, WAV vs MP3, and ends with a free path that needs no credit card.

The 30-second version

  1. Sign up free at tts.audio/sign-up. You get 10,000 characters a month with no card.
  2. Open the studio, paste your text into the composer, or drop in a PDF.
  3. Pick a voice from the dropdown.
  4. Click Generate.
  5. When the audio is ready, the player loads inline. The same screen gives you a download link to the MP3.

Short text finishes in one to three seconds. A 100,000-character document (about a 100-page book) takes 30 to 60 seconds because it is split into chunks and processed by a background worker.

If you just want to listen, not download

Skip the sign-up entirely. The free TTS audio page lets you type up to 300 characters, pick any voice, and hear it spoken instantly. Plays are unlimited. Downloads require an account because of compute cost.

PDF to MP3

tts.audio extracts text from PDFs using unpdf, splits it at sentence boundaries into 4,096-character chunks, and sends each chunk to the TTS engine. The resulting MP3s are concatenated losslessly with ffmpeg -c copy, so there is no re-encoding and no quality loss at the join points.

To convert a PDF:

  1. Open the studio.
  2. Click the upload area or drag a PDF onto it.
  3. Pick a voice.
  4. Click Generate.
  5. The page polls every two seconds. When the worker finishes, the player loads.

Supported file types beyond PDF: DOCX, EPUB, TXT, Markdown, HTML, RTF, CSV, JSON. Each is parsed into plain text and then treated the same way.

MP3 vs WAV

MP3 is the right default. It plays everywhere, the file size is small (about 1 MB per minute at standard TTS bitrate), and the quality is indistinguishable from the source for spoken voice.

WAV is uncompressed PCM. tts.audio serves WAV automatically when you use Gemini or Qwen voices, because those engines return raw PCM. WAV files are roughly 10x larger than MP3 for the same audio. Convert WAV to MP3 with ffmpeg -i input.wav -codec:a libmp3lame -qscale:a 2 output.mp3 if you need the smaller file.

Choosing a voice

The voices page lists 76 neural voices across 17+ languages. Each has a playable sample so you can hear the timbre before committing. MAI Voice 2 voices carry expressive styles (cheerful, sad, whispering, and so on); Gemini voices take inline mood tags like [whispers] and [laughs].

For audiobook-style narration, the en-US Harper and en-US Olivia voices are good neutral starting points. For podcast intros, try en-US Ethan or en-US Jasper. All are free to test on the free tool page.

What it costs

Free tier: 10,000 characters per month, no card. Paid starts at $0.13 per day for 100,000 characters on Pro, and $2.63 per day for 2,000,000 characters on Studio. Commercial usage rights are included on every paid plan. The free tier is personal-use only. See the pricing section for the full breakdown.

If you need this in code

tts.audio runs on top of OpenRouter, which exposes the same neural voices over a standard HTTP API. If you are building your own app and want to call the engines directly, point your client at https://openrouter.ai/api/v1/audio/speech with the model id (microsoft/mai-voice-2, google/gemini-3.1-flash-tts-preview, or qwen/qwen-audio-3.0-tts-plus) and a voice id from the voices catalog. The response is an audio URL or raw PCM depending on engine.

Related