Free AI Text to Speech with Voice Cloning

Instructions

Choose the language and the voice

In AI Voice Model, choose the language the narrator speaks. Each language is its own model: the first time you use one it is downloaded from this website (about 175 MB, 402 MB for French) and your browser keeps it, so later visits start without downloading it again. Then pick one of the voices published for that language.

Clone your own voice

Press Clone My Voice, choose a microphone and read the text shown out loud for 6 to 20 seconds, somewhere quiet. You can also use an existing recording file instead of the microphone. Listen to it, confirm that it is your voice or that you have the speaker's permission, and press Learn my voice. The first time this also downloads the voice encoder (39 MB).

Your voice is stored in this browser only, and it is the same voice the Flowchart Editor uses for its narrations. Once you have it, the built-in voices are off; delete it with the bin to choose one of them again.

Generate and download the narration

Type up to 50,000 characters in Text to Speech and press Generate voice (or Ctrl+Enter). Play it to listen, delete it with the bin to try again, and press Download MP3 to save it. The file is named after the first words of the text.

Open the gear beside optimized generation. Silence around the narration adds up to one second of silence before and up to one second after the narration, with either generate button, for example to give a video or a podcast a moment before the voice starts. Changing it updates a narration you already generated, without generating the voice again.

Generate voice optimized

Generate voice optimized keeps sentences together so commas have natural intonation. Long sentences are divided into smaller sections, preferably at clause boundaries. Blank lines separate paragraphs with longer pauses; a single line break simply continues the sentence. The voice gets extra time to finish word endings, and volume is balanced gently between sections.

The gear sets the pauses between sentences and long clauses, and can reduce extra silence while preserving quiet sounds and breaths. Silence already in the audio counts toward each pause. Texts over 2,000 characters use this processing with either generate button. Progress shows the current section; stopping keeps your previous complete narration and does not offer an unfinished file for download.

Narration speed, in the gear, defaults to 0.85× for a calmer pace. Adjust it from 0.75× to 1.10× and generate again. The voice keeps its pitch, and the speed is included in the downloaded MP3 as well as playback. Dense sentences get breathing room at substantial clause boundaries, without breaking short phrases at every comma.

Generation controls

Controls, beside the generate buttons, opens the voice model's settings. They change how the speech is generated and apply to the next narration. temperature sets how varied the delivery is, and sampler_decode_steps trades speed for quality. Optimized generation protects sentence endings even when saved end-of-speech settings are too short. Restore quality defaults resets the model controls. These settings, the optimized generation's and the silence are all kept in this browser for your next visit.

Credits

Pocket TTS © Kyutai, CC BY 4.0. ONNX export by pocket-tts-onnx, CC BY 4.0. Clone only your own voice or one you have permission to use.