chrisjp
commented
1 year ago All English language TikTok voices are now found. Other languages currently in progress.
Today I was able to figure out the request for fetching voices list JSON for each language in TikTok (with thanks to several people in the linked issue in the OP).
Bad news it requires an access_key which as far as I can tell is generated inside the app somewhere, so use of HTTP Toolkit and a modified TikTok .apk is required to obtain this.
Here's some quick and dirty documentation for it based on my testing...
API endpoint
GET https://api31-normal-useast2a.tiktokv.com/effect/api/v3/effects
Required parameters
access_key- 32 char alphanumeric string, looks like it might be an md5 hash. Seems to be generated in the app. No cookies generated by the website contain this value. You'll need HTTP Toolkit to find this.app_version- self-explanatory.30.1.2is the value I'm using, matches the modified .apksdk_version- unsure which SDK this refers to, possibly something internal.14.3.0is a working value.device_id- a 19-digit numeric[0-9]ID. You can fake this.1234567890123456789will work.device_platform- Must beandroid. I didn't find any other working values.device_type- model ID of your device. My emulator usesSM-G988N. You can probably put anything you want here.channel- Presumably the store the app was acquired from.googleplayworks here.panel- Panel is what TikTok calls each of its effect categories or something. You need to putspeaking-voicefor this.
Optional parameters
region - 2-letter uppercase country code. Defaults to US if not set.
carrier_region - 2-letter uppercase country code. Defaults to US if not set. On your device this will be the country it's physically located in based on SIM card network you're connected to.
Full URL example
https://api31-normal-useast2a.tiktokv.com/effect/api/v3/effects?access_key=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX&app_version=30.1.2&sdk_version=14.3.0&device_id=1234567890123456789&device_platform=android&device_type=SM-G988N&channel=googleplay®ion=US&carrier_region=US&panel=speaking-voice
you'd need a valid access_key in that for this to work, but no special headers need to be set. You could open it in your browser if you wanted to.
thetoonsofjosh
PeeFiftee
WaleedAnimate2020
Daniel49332
GJJellyfish
Last updated: 3rd October, 2023
A number of the APIs have voices available that we either don't know the IDs for (if closed source like TikTok for example) or we do know them but for some reason they don't work. This issue aims to keep track of them all and can be updated as and when they're resolved.
Included are the JSON entries that would be in
voices.jsonif they were working, in case you want to try them out for yourself on your own locally hosted copy of this library.TikTok
As a closed source app it's difficult to obtain the voice IDs. I previously relied on contributors to other GitHub repos to find these out. In the app there are a lot of named TTS voices that don't match their IDs, furthermore, there are many unique to certain countries/languages.
oscie57 and other contributors to https://github.com/oscie57/tiktok-voice/issues/1 have been a reliable source of voice IDs so far. And thanks to the input of others there I've been able able to ascertain all of the English language voices that were missing.
Currently unknown English language voice IDs: None!
It is now possible to obtain voice IDs from other countries (see replies), so these are currently in progress.
StreamElements
5 of the 'standard' engine Polly voices do not work. My guess is that these were intentionally removed by StreamElements from their API and replaced with better (in their opinion) alternatives from the Google Text-to-Speech ones they added. Additionally, the Hebrew voice "Doron" (not a Polly voice) returns an error response despite being selectable by streamers in their dashboard - I would expect StreamElements to either fix or remove this one at some point as it's their only option for Hebrew speakers.
StreamElements
No known issues.
CereProc
No known issues.
IBM Watson
No known issues.
Acapela
No known issues.
Oddcast
No known issues.
Google Translate
No known issues, however the translate site will need to be checked every so often to see if any languages have had a voice added.
Microsoft Azure Speech (via Bing Translator)
A grand total of 142 of Microsoft's bank of voices do not work via the Bing Translator API endpoint we use to generate audio. While some of these are legitimately still in what they call a 'preview' phase, the majority don't appear to have any limitations. My guess is that Microsoft feel these ones aren't as suitable for reading user translations (unclear accent maybe?), though in any case it's odd that Bing specifically isn't able to access them given it's a first party product of theirs. I don't hold much hope for many of these to ever work though those in preview (with a superscript ¹) may be worth checking from time to time.
VoiceForge
One voice always results in a timeout error.
ReadSpeaker
Removed. Demo website is now paywalled and voice synthesis can no longer be accessed for free.
iSpeech
Removed. Demo website has been non-functional since late 2022. The API is still functioning, though, but obviously requires a valid API key. As such this has effectively paywalled the service.Now available again (as of 3rd October, 2023) by using their actual API this time.The "Greek Female" voice is no longer listed in the documentation but is still working. The "Taiwan Chinese Female" voice works with the demo key but errors with some API keys, reporting it as an "Invalid voice". It is unknown why this happens.