Skip to content

Latest commit

 

History

History
84 lines (67 loc) · 6.14 KB

README.md

File metadata and controls

84 lines (67 loc) · 6.14 KB
Raw

Text-to-Speech (TTS) Demo App

A simple web app demonstrating how text sounds in different TTS voices.

Convert your text to speech using voices provided by a variety of TTS services, the full list including their limitations can be found further down this readme.

About

This project originally came about from a desire to emulate how messages will sound when donating to Twitch.tv streamers, many of whom use Streamlabs/StreamElements TTS features to have donations read aloud (often by 'Brian') during their streams. FeelsGoodMan

It was inspired by Peter Cunha's similar tool. I wanted to build it using only vanilla JS, as well as improve it with UI/UX enhancements and by providing a massive selection of voices, including all those offered by Amazon Polly and other services. The project has grown somewhat bigger than the single file demo it was initially, and is now effectively a full PHP library and demo website.

Features

  • Over 1000 different high-quality voices
  • Lightning fast
  • Copy link to TTS URL to clipboard for sharing with others
  • Save audio locally for a configurable amount of time
  • Plain old HTML5 and vanilla JavaScript (PHP >=8 required for full functionality)
  • Conversations - create a conversation using multiple TTS voices and share it as a playlist of TTS audio clips.

Using

The basic requirements are some kind of web server with PHP >= 8.0 installed and a working copy of composer. A Docker compose file is included to handle all this for you.

docker compose up

If not using Docker I'll assume you've already got this repo on a server; you'll need to run

composer install

in the directory to generate the autoloader classes.

Please check the generated config.php file and adjust any values as you see fit. Take note of the step to acquire a valid TikTok session ID value if you intend to use those voices, as these will not generate any audio without one. Lastly, ensure the AUDIO_DIR exists and is writable, especially if you have changed it from the default value. Composer will have attempted to create the default directory for you, and the script itself will always try to create your configured directory if it does not exist. But depending on how your PHP is set up this may not be possible.

Usage notes

Many of the services used aren't intended for professional/commercial use or are utilising a live demo, and as such they have limitations. As a reminder: You should only be using this to test how your text sounds in various voices. If you need to use TTS for commercial purposes please check out the services' respective websites linked below for more information.

Regarding URL validity, it is recommended you enable local saving in config.php if you'll be requiring access to the audio again in the near future. The same advice goes for services that return the audio as a data URI, as this can be inconvenient for sharing when message length is a concern.

TTS Services/APIs included (and their limitations)

Amazon Polly (via StreamElements or Streamlabs APIs)

  • (StreamElements) 3000 byte text limit
  • (StreamElements) Generated audio (MP3) is returned as a data URI
  • (Streamlabs) 550 byte text limit
  • (Streamlabs) TTS URLs are valid for only 5 minutes

Google Cloud Text-to-Speech (via StreamElements API)

  • 3000 byte text limit

CereProc

  • 2000 byte text limit

TikTok

  • 300 character text limit
  • valid session ID cookie required (see config.php for details on how to get one)
  • API is undocumented and not all voice IDs are known
  • Generated audio (MP3) is returned as a data URI

IBM Watson

  • Generated audio (MP3) is returned as a data URI

Cerence

  • Generated audio (MP3) is returned as a data URI
  • text limit was unable to be accurately determined and seemed to depend on numerous factors, including voice and/or language. In testing I was able to get above 1900 characters with many voices, but certain strings (e.g. a long line of emojis) would result in an error even if the length was well under that. Text of an equivalent length in languages that do not use a Latin character set would also produce an error. Limit is therefore set at 1900 bytes as this should work for most voices.

Acapela

  • 2000 character text limit

Oddcast

  • 600 character text limit

VoiceForge

  • Generated audio (WAV) is returned as a data URI
  • 540 character text limit

Cepstral

  • No practical character limit appears to be in place, but it is set to 5000 because anything over this takes well over a minute to generate.
  • Their server rate-limits requests. It is highly likely that Cepstral will often not work on the demo due to the traffic it receives. It is recommended you use their own site directly, otherwise you'll have to host this library yourself.

Microsoft Azure Speech (via Bing Translator)

  • 3000 byte text limit
  • Generated audio (MP3) is returned as a data URI

Google Translate

  • 200 character limit
  • Generated audio (MP3) is returned as a data URI

iSpeech

  • Requires a valid API key as their public demo website is no longer functional. A demo key for evaluation purposes is provided if you do not have one.
  • Depending on the permissions of the API key being used, TTS will be limited to a variable word count (maximum of 100 for the demo key).
  • Some non-English voices cannot interpret English text and will instead say each individual letter.
  • API server can be unreliable and sometimes not return a response.
  • Generated audio (MP3) is returned as a data URI

Demo website

Head over to LazyPy.ro/tts for a live demo of all the available voices.