Ivona Text To Speech - Doesn't work with French accent

[LapinFou]

Hi Caffeine,

I got an issue with Ivona TTS. I don't know where is the problem, but if there is some French accent, the MP3 is not generated. Here is an example: telemko | télémétrie perdue I'm using the voice Céline, fr-FR, Female

I hope you can fix this issue, because it is a very good voice.

[CaffeineAU]

Hi Sebastien. the issue lies with getting Ivona to accept accented characters. Their documentation specifies that UTF8 encoded accented characters are accepted, but I haven't been able to get that to work yet. I'll investigate more.

[LapinFou]

OK. Thanks. If I go directly on https://www.ivona.com/ it does work.

[CaffeineAU]

I have resolved the issue. When you UTF8 encode the string, the length changes because of the double width of certain characters. I wasn't taking the length change in to account when calculating the payload length to send to Ivona.

I'll generate a release now.

[CaffeineAU]

Changes incorporated in to release 2.8.2.1

[LapinFou]

I tried. It is better (doesn't stuck), but it is still not working perfectly. It seems that "é è" are replaced with a "e". This is not correct, because the pronunciation is not correct at all.

[CaffeineAU]

So it sounds different to when you put the text in to the Ivona webpage directly?

[CaffeineAU]

Here are two samples, of audio. One is created with the example text you gave (télémétrie perdue) and the other is the same without the accents (telemetrie perdue)

They sound very different to me. I think the accented one sounds almost the same as directly putting it on to the Ivona webpage, but not being a native French speaker, I can't tell how wrong it is.

Samples.zip

[LapinFou]

Hi Caffeine,

Sorry for the late answer. I was away from home during few days. When I copy/paste this sentence: télémétrie retrouvée on the Ivona website, it is perfect. If I used, TTSAutomate, it is not correct. Even for a none French speaker, you can hear a big difference when using the French, Mathieu voice.

Are you sure, that you did not convert the accent é (hex code c3 a9) in è (hex code c3 a8) ? Because It is not the same at all. télémétrie retrouvée (correct) gives the same voice than tèlèmètrie retrouvèe (not correct).

I think, this is the problem.

[CaffeineAU]

Hi Sebastien,

I have resolved the issue. Please try 2.8.2.2

[LapinFou]

I will give a try ASAP. I will keep you inform. Thank again for your work. 😄

[LapinFou]

I tried. It is working !!! Ivona is definitely one of the best French voice.

FYI, the following sentence is not working: vingt & un (21) in French, we pronounced "twenty AND one". It is working on the Ivona website. but vingt et un does work. Probably the character & is not handled.

[CaffeineAU]

Thanks for testing this. I'll look into the handling of the ampersand.

[LapinFou]

You got 2 solutions:

1. handle the ampersand (nice one)
2. forbid the usage of the ampersand (quick one). 😺

[LapinFou]

You got 2 solutions:

1. handle the ampersand (nice one)
2. forbid the usage of the ampersand (quick one). 😺

[CaffeineAU]

I think handling it is best, although if I replace it with 'and', that won't work for 'et'

I'll see how Ivona is doing it

[CaffeineAU]

Wrong button, sorry!

[CaffeineAU]

I have identified that escaping the ampersand with & amp; resolves the issue. I'll create a release for this shortly.

[CaffeineAU]

The issue is resolved in 2.8.2.4

It's addressed in Ivona and Microsoft engines. I haven't checked whether Google is susceptible yet.

[LapinFou]

I checked in Ivona and Microsoft engines. It is working. Doesn't work with Google.

[CaffeineAU]

Google requires a separate approach; instead of replacing & with & amp;, it must be replaced with %26

I have done this and will now publish 2.8.2.5

[LapinFou]

It does work. Well done! Thanks again. 😄