Forum Discussion
A few questions about Polish language support in t2s
Hi we're heavy users of the text 2 speech capability in Polish language. After the new voices have been introduced the quality of the speech produced is much better. I have a few questions:
1. What is the roadmap for Polish neural voices? I am asking here about
- more male voices (and kids),
- support for voice styles in Polish,
2. Are there any plans to support what amazon calls "specifying another language for specific words", that is, the ability to force t2s service to read e.g. english words using polish voice.
3. How about phonemes for Polish language ?
4. The Polish version of the voice we're using is a bit buggy. Some words get mispronounced, which requires us to re-write texts. Is there any fast escalation path that would allow us to contact you directly to solve the issue?
5. We're also experimenting with speech 2 text, however what we're really interested in is a scenario with speaker recognition. Any plans to implement that for Polish language?
6 Replies
- Hi Everyone, is there any update on #5? Are there any plans to support Polish language for speaker recognition?
- edwardun
Microsoft
Thanks macieii for the questions.
1. They are in the roadmap. Also, supporting Polish is also our roadmap for Custom Neural Voice. You can build more customized styles and voices soon.
3. Does the current phonemes support in IPA not adequate? Can you let us know what is missing?
4. You can send us reports through the following channel: https://docs.microsoft.com/en-us/azure/cognitive-services/cognitive-services-support-options?context=/azure/cognitive-services/speech-service/context/context
- Okay, thank you. #1 - can you reveal any dates?
#3 - I will use the channel provided
#4 - thank you.
pranavpanditmacieii for #2, does this help? Speech Synthesis Markup Language (SSML) - Speech service - Azure Cognitive Services | Microsoft Docs
You can specify multiple voices in your SSML for specific segments.
You may also want to look at the say-as and interpret-as and phoneme elements here: Speech Synthesis Markup Language (SSML) - Speech service - Azure Cognitive Services | Microsoft Docs- Re the first part of the answer. The idea is to use the same voice. Amazon Polly uses a dedicated ssml tag <lang xml:lang="fr-FR">Je ne parle pas français.</lang>. Now suppose the original voice is set to Polish (for example). The outcome of this fragment is the Polish voice trying to pronounce the French phrase. The concept is quite natural, it sounds as a Polish native speaker was trying to speech French.
Say-as and interpret-as are a different story. The phonemes ... are pretty effort-consuming in preparation. That's why I consider the amazon's <lang> element a nice feature.
The most frequent case: Polish voice reading English names, or Spanish voice reading Polish names.
