Bug Report: Mispronunciation of Welsh Contraction "i’w" in Azure Neural TTS

Question

Subject: Bug Report: Mispronunciation of Welsh Contraction "i’w" in Azure Neural TTS

Description:

The Azure Neural TTS system is mispronouncing the Welsh contraction "i’w." Instead of producing the correct pronunciation ("ee-oo" or "ee-uh"), the system reads it as two separate sounds, "i w." This issue affects all Welsh language voices.

Steps to Reproduce:

1. Input Text:

• Use any Welsh language input text containing the contraction "i’w." For example: "Rydw i'n mynd i’w gweld yfory."

2. Select Voice:

• Choose any Welsh language voice in the Azure Neural TTS system.

3. Generate Speech:

• Synthesize the input text to speech.

Expected Result:

The contraction "i’w" should be pronounced as "ee-oo" or "ee-uh," producing a smooth, combined sound.

Actual Result:

The system pronounces "i’w" as two separate sounds, "i w," which is incorrect and disrupts the natural flow of the sentence.

Additional Notes:

• The issue is consistent across all Welsh voices in the Azure Neural TTS system.
• Alternative spellings or representations (e.g., "iw," "i'w," or "i 'w") were tested but did not resolve the issue.

What is the proper channel to report this through? Does annyone have ideas to mitigate it? Ideally this would be addressed in the next version of the model to improve the accuracy and naturalness of Welsh language synthesis.

Thank you for your attention to this matter.

Answer 1

@Verbari LLC Thanks for checking our audio file and confirming that the text to speech for Welsh is working as expected.

.

We will be deploying this fix by end of this June month. Hope this helps.

**

Please do not forget to "Accept the answer” and “up-vote” wherever the information provided helps you, this can be beneficial to other community members.