Polly voice "Amy" pronunciation & lexicon issues

Hi TB,

Thanks for your reply. That's correct - we have long had a lexicon file uploaded which has been referenced using the api by the Amy Polly voice for many years now. Over the last months, as AWS changed the Amy voice, this referencing had been ignored which meant all the pronunciations in the lexicon file were ignored, vastly increasing our workload and error checking. However, I'm pleased to report that over the last week, it appears that the lexicon file is again being referenced correctly. I haven't had notifications of a fix, but all appears to be working ok now.

I shall keep monitoring and return to this post if problem arise again.

It would be helpful to know whether AWS recognised the problem and implemented a fix.

thanks, Nick

To ensure your lexicon file is applied during synthesis, you need to explicitly reference it in your API call to the SynthesizeSpeech operation. Lexicons are stored in specific AWS regions and must be specified when synthesizing text - they are not automatically applied.

When you call the SynthesizeSpeech API, you should include the lexicon name(s) in your request parameters. You can apply one or more lexicons from your region during the synthesis process. The lexicon will be applied to the input text before synthesis begins.

For the pronunciation issues you're experiencing with words like "Well-regarded" and "UPVC," as well as place names, you can address these by adding entries to your lexicon file. Lexicons give you control over how Amazon Polly pronounces words, including:

1. Stylized words or acronyms: You can define aliases for words that should be pronounced differently than they're spelled (like "UPVC")

2. Phonetic pronunciation: You can specify exact pronunciations using phonetic alphabets (IPA or X-SAMPA) for words that aren't being pronounced correctly

3. Custom pronunciations: For place names or hyphenated words that are being mispronounced, you can create lexicon entries that specify the correct pronunciation

Your lexicon file should be W3C PLS-compliant XML format. For each problematic word, you can create a lexeme entry that either provides an alias (how it should be read) or specifies the phonetic pronunciation.

Make sure you're using the PutLexicon operation to store or update your lexicon in the correct AWS region, and then reference that lexicon name in your SynthesizeSpeech API calls to ensure it's being applied consistently.
Sources
Managing lexicons - Amazon Polly
Use PutLexicon with an AWS SDK or CLI - Amazon Polly

Thanks, but as I mentioned, we've been supplying and Amy has been using the lexicon file for years without a problem. We have always called it via the api - so unless you have changed your methods in the last week(?) it should still work.

Yet we are finding that it is now not being referenced at all. Please help!

Thanks, Nick