Here at Nexmo we do use text-to-speech in our telephony applications extensively, but did you know that your text could do more than just speak? Our Voice API supports SSML (Speech Synthesis Markup Language) which allows you to add some expression into your text-to-speech outputs.
Since it’s Christmas, I thought I’d set myself a Christmas Poem as a challenge! Text-to-speech with just punctuation such as commas and full stops actually works pretty well, so even on the first attempt, you could sort of tell it was supposed to have meter.
Little Jack Horner
Sat in the corner,
Eating of Christmas pie:
He put in his thumb,
And pulled out a plum,
And said, “What a good boy am I!”
We can improve on this by marking up the words that should have emphasis. I used
<prosody> tags to give the strong words in the rhyme more volume and to slow them down. For example “Horner”, “corner”, “plum” and “thumb” are all marked up with
<prosody rate="x-slow"> to help with the rhythm of the speech.
The text-to-speech didn’t know the word “Horner” so I used a
<phoneme> tag to spell phonetically how the word should be pronounced (which at least makes it rhyme with “corner” in the next line, it still sounds a bit strange to me!). This is a very useful trick when curating your text-to-speech content, especially for proper nouns which are often unfamiliar to the parser. You can also use it for any other words whose pronunciation doesn’t come out as you expected.
Finally, for fun, I bleeped out the word “good” in the last line by making use of the
<say-as> tag and making that word an expletive. This tag can be very useful if your application needs to speak user-supplied content with unknown contents!
Here’s the SSML I ended up with:
Little Jack <prosody rate="x-slow"><phoneme alphabet="ipa" ph="ˈhɔːnə">Horner</phoneme></prosody>,
Sat in the<prosody rate="x-slow">corner</prosody>,
Eating of <prosody volume="x-loud">Christmas</prosody> <prosody rate="x-slow">pie:</prosody>
He put in his <prosody rate="x-slow">thumb</prosody>,
And pulled out a <prosody rate="x-slow">plum</prosody>,
And said, “What a <say-as interpret-as="expletive">good</say-as> boy am <prosody rate="x-slow">I</prosody>!”
By including this XML as the
text field in the
talk action of my NCCO, and adding a
record action too, I was able to capture the poetry of the robot:
You could consider adding some more expression to your spoken interaction with your users by adding SSML – and poetry is a great way to practice. Let us know if you build something poetic this Christmas!