Cloud Text-to-Speech Voices updating December in European markets

[joshuag]

I got an email in September titled: "Cloud Text-to-Speech voices will be updated starting December 6, 2024 across European markets"

It states that voices will be updated in December 2024 to newer versions.

Is there anywhere that allows the new voices to be previewed. We want to check for differences between the older and newer versions before the rollout starts in case we need to choose different voices.

[MarvinLlamas]

Hi @joshuag,

Welcome to Google Cloud Community!

To preview the new voices and assess them alongside the previous versions, you can refer to the 

Google Cloud Text-to-Speech documentation, where the release notes and supported voices and languages pages provide in-depth details on the updates and available voices.

Additionally, you may want to visit the Google Cloud Blog for any announcements or updates regarding Cloud Text-to-Speech, or you could use the voices:list API endpoint to retrieve a list of all available voices and compare their features. This will help you spot any differences and make well-informed choices about which voices to select.

I hope the above information is helpful.

[joshuag]

None of those links mention new voices for the EU though.

The email states:

> Starting December 6, 2024, the Text-to-Speech voices listed below will automatically transition to an updated version within the same tier. The new voices might not resemble the original speaker's voice, but they will be of higher quality and sound more natural.

My concern is that the "new voices might not resemble the original speaker's voice".

[ChrisTurner2TVS]

Hi - I noticed this week that the Neural 2 voices have disappeared and are no longer available. I have been using the en-GB-Neural2-C voice for a while now. Any ideas where the Neural 2 voices have gone? All I can see available now are the "new" en-GB-Chirp3 voices.

[joshuag]

There has been additional communication listing the voices that are being replaced and a link to a table: https://services.google.com/fh/files/emails/b_362612430_reminder_1.pdf

However, some of the new voices listed don't yet exist, but are being replaced at the end of January.

It's also highly disappointing that the choice is being reduced.

[pawelwaleczek]

Regarding the change, I'd like to clarify how it works.
Taking the first voice from the PDF as an example:
bg-BG-Standard-A => bg-BG-Standard-B

After Jan 27, 2025 can I still use the 'bg-BG-Standard-A' name in the API for voice synthesis and it will just sound like 'bg-BG-Standard-B' or I MUST to replace my API request to use 'bg-BG-Standard-B' voice name to avoid errors.

[joshuag]

The voices are apparently being replaced in 20 days, but the en-GB-Wavenet replacements don't yet exist.

For clarity, the PDF mentions en-GB-Wavenet-A (and C and F) are being replaced with en-GB-Wavenet-N and en-GB-Wavenet-B (and D) are being replaced with en-GB-Wavenet-O.
Neither en-GB-Wavenet-N or en-GB-Wavenet-O exist in the voice list on https://cloud.google.com/text-to-speech/docs/voices

Randomly searching through the PDF table, it looks like it-IT-Wavenet-E/F and nl-NL-Wavenet-F/G have the same problem

[MikeeSouth]

I'm also following this change and I'm following the Swedish voices. Apparently sv-SE-Wavenet-A, B, C, D and E (5 voices) will be replaced with F and G (2 voices). Unfortunately these voices are significantly worse than the current A-E voices.

@MarvinLlamas do you have any insight in this change? The replacement voices are worse for us and we need to either keep the old voices or we will need to change to another TTS provider. The new replacement voices for Swedish is more like Standard voices rather than Wavenet voices.

As @pawelwaleczek asked, I'm also wondering if we need to update our API calls to send the new letters or if we can still send the old letters and it will be automatically use the new replacement voices.

@joshuag, I see that you have several voices that don't exist yet but do you also experience that the new voices are worse than the current ones or do you find the new replacement voices at the same level or better as the current ones?

[nordholm]

I, too, would like clarity on whether we need to update our API calls to send the new letters or if we can still send the old letters and it will be automatically use the new replacement voices.

[Andy77548]

Ah glad it wasn't me! Yes I've noticed a significant degradation in the Wavenet Voices, or the list voices API is still returning an outdated list..the API query still returns voices A - F? In particular I notice that sv-SE-Wavenet-D sounds exactly like sv-SE-Standard-D, ditto with voice 'A', which suggests an error on Google's part?

[MikeeSouth]

Yeah, this is really messed up by Google unfortunately. Did you get it wrong when you wrote that sv-SE-Wavenet-D sounds exactly like sv-SE-Standard-D? They are not the same for me, but sv-SE-Wavenet-D sounds exactly like sv-SE-Standard-C.

They are replacing Wavenet-D with Wavenet-F (that sounds much worse). So a fallback would be to use Standard-C, right? Nope. Google is replacing Standard-C with Standard-F and guess what? Standard-F sounds exactly like Wavenet-F. So they are replacing the good voices with bad ones and making Standard and Wavenet have the same options and that they sound exactly the same. I don't get it.

I've tried in multiple channels to reach out to Google about this (and other problems with their0 TTS) but without success. Account managers, startup programs, support tickets, google groups, this community - nothing seem to reach them.

I'm currently migrating to Microsoft TTS.

[Andy77548]

Nice to know it wasn't some error in my code! Thanks

[mono-math]

Anyone know if there's a way of using the voices that have been replaced? I'm in the middle of a project and did not know these changes were going to happen!

[MikeeSouth]

@mono-math, the old voices are still available but were supposed to be replaced 2 days ago (27th of January). You can see the letters for the new replacement voices in this PDF: https://services.google.com/fh/files/emails/b_362612430_reminder_1.pdf

[sargearmstrong]

We've been using 'en-GB-WavenetA' but it sounds diffferent now. We tried various variants (as follow), but all sound alike and nothing like the samples indicate they should below. Seems they bugged something. 

https://cloud.google.com/static/text-to-speech/docs/audio/en-GB-Wavenet-N.wav (a new voice but not on sample page yet)
https://cloud.google.com/static/text-to-speech/docs/audio/en-GB-Neural2-N.wav (a new voice but not on sample page yet)

https://cloud.google.com/static/text-to-speech/docs/audio/en-GB-Neural2-A.wav (not accessible in API, sounds like something else)
https://cloud.google.com/static/text-to-speech/docs/audio/en-GB-Wavenet-A.wav (not accessible in API, sounds like something else)

[MikeeSouth]

@sargearmstrong thanks for pointing that out, we actually noticed the same thing yesterday. Google has replaced all the old voices with the new voices, without forcing us to actually change letters for the voices. So our voice "D" is now effectively the new voice "F" but we can still request voice D.

I'm no longer using Google TTS since the new voices are much worse for Swedish.

PS. I have no idea on "not accessible in API, sounds like something else". Good luck with that.

[mono-math]

En-GB-Wavenet-A did change two days ago. The voice is gone completely as an option. It’s also not in the “new replacement voice” column so there’s no indication that it has replaced anything else.

[joshuag]

Have the voices changed back?

en-GB-Wavenet sounds like more than two voices available.

[mono-math]

Seems to have happened again! This time with no notice at all, as far as I can see? All Wavenet and journey voices gone. Replaced with chirp. I need the wavenet voices to make changes to an existing project!

[ChrisTurner2TVS]

Me too. Only the Chirp voices available now.