IssueTracker

Problem you have encountered:

When generation text to speech files with the Chirp-HD voices, questions marks are usually used correctly, but around in 1/3 or 1/4 of cases the next few letters of the following word are said weirdly before starting to say that word correctly. Listen to the attached files for as an example.

What you expected to happen:

To not have artifacts after questions.

Steps to reproduce:

1. Go to https://console.cloud.google.com/speech/text-to-speech
2. Select en-US-Chip-HD-D
3. Copy the following example: Considering your interest in Art Nouveau and your background as a Computer Science student, are you perhaps interested in how digital tools can be used to analyze and preserve architectural heritage? Or do you rather want a more classical approach to architecture?......
4. Generate the sound file multiple times until you hear the artifact after the question (see attached files)

Other information (workarounds you have tried, documentation consulted, etc):

While the documentations mentions that weird artifact like this can happen, this issue seems to be reproducible and might be an underlying issue.