Azure TTS: Synthesis of single word breaks in the middle of the word

DWinkler 26

We are synthesizing with SSML the word dieci / number 10.

The synthesis only says /d i e/ and doesn't give the whole word.

If we synthesize 'diece minuti' it works perfectly.

e.g.

<speak xmlns=\"http://www.w3.org/2001/10/synthesis\" xmlns:mstts=\"http://www.w3.org/2001/mstts\" version=\"1.0\" xml:lang=\"it-IT\"><voice name=\"it-IT-FabiolaNeural\"><prosody rate=\"0.87\">dieci</prosody></voice></speak>

This is happening for a few voices, it-IT-ElsaNeural is fine on the other hand

i also tried the say as cardinal, which has the same problem

providing the phonemes synthesizes the whole word, but sounds wrong /ˈdjɛt͡ʃi/

YutongTie-MSFT 51,776 Reputation points

2024-09-19T01:51:48.1766667+00:00

Hello @DWinkler

Thanks for reaching out to us, could you please share which voices you saw this issue so that we can investigate it from our end.

Regards,

Yutong
DWinkler 26 Reputation points

2024-09-19T06:29:22.7466667+00:00
Thanks for getting back to me so quickly!

I tried a few and here are the results:

Good:

ElsaNeural

Stop after /d i e/:

FabiolaNeural

IsabellaNeural

CalimeroNeural

FiammaNeural

GianniNeural

PierinaNeural

Bad synthesis

DiegoNeural (/d i e x i/) but "dieci minuti" is fine

BenignoNeural
YutongTie-MSFT 51,776 Reputation points

2024-09-23T21:38:00.79+00:00

Hello @DWinkler

Thanks for your information again, want to update to you that we have notified product team for investigate it. Will let you know once we get any news from product team.

Regards,

Yutong
DWinkler 26 Reputation points

2024-09-24T11:30:12.6533333+00:00

Thank you for keeping me posted @YutongTie-MSFT

I would like to add that we come across this issue in different languages too, although i cannot give such a detailed information as for the italian case above. Other languages where we observed it are: es-ES, de-AT, it-IT, de-DE

Share via

Azure TTS: Synthesis of single word breaks in the middle of the word

Your answer