Text to Speech Pro

How do I turn text into speech with multiple voices and download the audio?

Write a script in AnyTool Text to Speech Pro, starting each line with a speaker name like “Alice:”, and give every speaker a different voice. Add per-line speed and pitch, pauses such as [pause 500ms], and a pronunciation dictionary, then press Play. To get a downloadable file, switch to the neural voice, load the on-device model once, and render your script to a WAV. Everything runs in your browser, so your text is never uploaded.

Multi-voice conversation scripts — assign a different system voice per speaker
Per-line {rate:1.2 pitch:0.9} overrides and [pause 500ms] / ellipsis pause markup
Pronunciation dictionary to fix names, brands and acronyms (say GIF as jiff)
Optional on-device neural voice renders downloadable WAV audio offline
Script and settings auto-save locally — 100% client-side, no upload

What is

Text to Speech Pro is an advanced in-browser text-to-speech studio that reads multi-speaker scripts aloud with a different voice per speaker, supports per-line speed, pitch and pause markup and a pronunciation dictionary, and can optionally run an on-device neural model to export a downloadable audio file — all without uploading the text.

Text Tools

Related terms

Text to SpeechNeural TTSVoice GeneratorSpeech SynthesisSSML

Frequently Asked Questions

The instant system voices use the Web Speech API, which can play audio but cannot save it to a file. To download audio, switch to the neural voice, load the small on-device model once, then render your script to a WAV file you can save and reuse. The model runs in a background worker in your browser, so the text never leaves your device.

Yes, with the neural voice. Browser system voices play live but cannot be exported; the optional on-device neural model renders a downloadable WAV.

Write your text like a screenplay: begin each line with a speaker label and a colon, for example “Alice: Hello” and “Bob: Hi there”. Text to Speech Pro detects each speaker and lets you assign any voice installed on your device to each one, so the conversation is read back in different voices in order, with the active line highlighted.

Start each line with a name and a colon, like “Alice:”, then pick a voice for that speaker in the Voices panel.

The neural voice is entirely optional. The page loads nothing heavy on open — the model is fetched only when you click Load neural voice, after which it is cached and works offline. Both the system and neural engines run inside your browser with no server, no account and no tracking, so your scripts stay private.

No, it is strictly opt-in. Nothing heavy loads until you click Load neural voice, and even then everything runs on your device.

Insert a pause anywhere with markup like [pause 500ms], [pause 1s] or simply an ellipsis. To correct pronunciation, open the pronunciation dictionary and add a rule such as say “Nguyen” as “win” or say “GIF” as “jiff”; the substitution is applied to both the system and neural engines before speaking.

Type [pause 500ms] or an ellipsis to insert a pause, and add a say-as rule in the pronunciation dictionary to respell a word.

Detailed Explanation

Methodology

How Text to Speech Pro Works

Text to Speech Pro offers two complementary engines that both run entirely in the browser. The system-voice engine drives the Web Speech API (window.speechSynthesis): it parses a conversation script into ordered, speaker-tagged blocks, assigns a different installed voice to each speaker, applies per-line speed and pitch multipliers and pause markup, and queues utterances block by block. The neural engine is an opt-in on-device model (SpeechT5 with a HiFi-GAN vocoder, run via transformers.js and ONNX Runtime Web in a Web Worker) that produces raw 16 kHz PCM, which the tool encodes to a downloadable 16-bit WAV.

Script syntax: lines like “Alice:” map text to speakers; {rate:1.2 pitch:0.9} sets per-line prosody
Pause markup [pause 500ms], [pause 1s] and ellipses insert timed silence
A pronunciation dictionary applies whole-word, case-insensitive say-as substitutions
System voices play live via the Web Speech API but cannot be exported to a file
The neural model is fetched only on explicit opt-in, then runs offline in a worker

Use Cases

Difference from the Standard Text to Speech Tool

The standard Text to Speech tool is a single-voice reader built on the Web Speech API for instant, private reading aloud. Text to Speech Pro adds multi-speaker scripts with one voice per speaker, per-line prosody overrides, SSML-lite pause markup, a pronunciation override dictionary, persistent saved scripts, and — uniquely — an optional neural voice that exports downloadable WAV audio. Producing a saved audio file is the key capability the Web Speech API alone cannot provide.

Standard tool: one voice, instant read-aloud, no audio file
Pro tool: multiple voices per script, prosody and pause control
Pro tool adds a pronunciation dictionary and saved scripts
Only the Pro neural engine can export a downloadable audio file

Privacy & Security

Privacy and Offline Use

Both engines run inside the browser, so the script is never uploaded, there is no account and no tracking, and the script and settings are stored only in local storage on the device. The neural model loads nothing on page open; after a one-time download it is cached and synthesizes offline. This contrasts with cloud TTS services, which send text to a server for processing.

Text to Speech Pro (in-browser) vs cloud TTS services
Capability	AnyTool TTS Pro	Cloud TTS (NaturalReader, Murf, ElevenLabs, Speechify, Play.ht)
Where text is processed	In your browser	Sent to the provider servers
Privacy	Text never leaves device	Text uploaded for processing
Multi-voice scripts	Yes, a voice per speaker	Varies; often higher tiers
Downloadable audio	Yes, WAV via on-device neural voice	Yes, but typically paid or capped
Works offline	Yes, after a one-time model download	No
Cost / signup	Free, no signup	Mostly subscription or credit-based
Voice realism	Good (system + small neural model)	Often higher with large cloud models

Cloud services such as ElevenLabs and Murf can offer more lifelike large-model voices, but they upload text and are paid; AnyTool processes everything locally and is free. Capabilities reflect the live AnyTool tool; competitor behaviour varies by plan.