Amazon Polly is a service that turns text into lifelike speech. Amazon Polly lets you create applications that can talk, enabling you to create entirely new categories of speech-enabled products. In this webinar, you’ll get an overview of how Polly uses advanced deep learning technologies to synthesize speech that sounds like a human voice. You’ll also learn how you can use Polly’s 47 lifelike voices and support for 24 languages to build speech-enabled applications that work in many different countries. Learning Objectives: • Learn about the capabilities and features of Amazon Polly • Learn about the benefits of Amazon Polly • Learn about the different use cases • Learn how to get started using Amazon Polly • Learn how Polly speech audio can be distributed without restriction • Get an overview of SSML • Understand what is included in the AWS Free Tier and how to estimate usage costs

1. © 2016, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Rafal Kuklinski, Piotr Lewalski
Amazon Text-to-Speech
12/09/2016
Amazon Polly
A service that turns text into lifelike speech

2.  Introduction to Amazon Polly
 Features and functionalities
 Text-to-Speech: Under the Hood
 Getting started
 Workshop & Demo
 Pricing
 Q&A
What to Expect from the Session

3. Introduction to Amazon Polly

4. Why we built Polly
 Apps using voice to communicate with end-users are
becoming more common every day
 Naturalness of generated speech is a key element of
user experience
 Integration of speech varies across use cases

5. What is Polly
 A service that converts text into lifelike speech
 Offers 47 lifelike voices and 24 languages
 Low latency responses enable developers to build real-time systems
 Developers can store, replay and distribute generated speech

6. Polly – Quality
Natural sounding speech
A subjective measure of how close TTS output is to human speech.
Accurate text processing
Ability of the system to interpret common text formats such as abbreviations, numerical
sequences, homographs etc.
Today in Las Vegas, NV it's 90°F.
"We live for the music", live from the Madison Square Garden.
Highly intelligibile
A measure of how comprehensible speech is.
”Peter Piper picked a peck of pickled peppers.”

7. Polly – Language Portfolio
Americas:
 Brazilian Portuguese
 Canadian French
 English (US)
 Spanish (US)
A-PAC:
 Australian English
 Indian English
 Japanese
EMEA:
 Danish
 Dutch
 British English
 French
 German
 Icelandic
 Italian
 Norwegian
 Polish
 Portuguese
 Romanian
 Russian
 Spanish
 Swedish
 Turkish
 Welsh
 Welsh English

8. Features and Functionality

9. Polly features: SSML
Speech Synthesis Markup Language
is a W3C recommendation, an XML-based markup language for speech
synthesis applications
<speak>
My name is Kuklinski. It is spelled
<prosody rate='x-slow'>
<say-as interpret-as="characters">Kuklinski</say-as>
</prosody>
</speak>

10. Polly features: Lexicons
Enables developers to customize the pronunciation of
words or phrases
My daughter’s name is Kaja.
<lexeme>
<grapheme>Kaja</grapheme>
<grapheme>kaja</grapheme>
<grapheme>KAJA</grapheme>
<phoneme>"kaI.@</phoneme>
</lexeme>

11. Text-to-Speech: Under the Hood

12. Goal: Convert text into intelligible, accurate, and natural speech
Challenges
• Homographs: words written identically that have different
pronunciation
I live in Las Vegas vs This presentation broadcasts live from Las Vegas
• Text normalization: disambiguation of abbreviations, acronyms, units
‘St.’ expanded as ‘street’ or ‘saint’
• Conversion of text to phonemes (Grapheme-to-Phoneme) in
languages with complex mapping such as English e.g. tough,
through, though
• Foreign words (déjà vu), proper names (François Hollande), slang
(ASAP, LOL) etc.
Main Challenges of Text-to-Speech

13. TEXT
Market grew by > 20%.
WORDSPHONEMES
{
{
{
{
{
ˈtwɛn.ti
pɚ.ˈsɛnt
ˈmɑɹ.kət ˈgɹu baɪ ˈmoʊɹ
ˈðæn
PROSODY CONTOURUNIT SELECTION AND ADAPTATION
TEXT PROCESSING
PROSODY MODIFICATIONSTREAMING
Market grew by more
than
twenty
percent
Speech units
inventory

14. Unit Selection
Conversion of phoneme sequence to waveform
Database of recorded audio
Unit – diphone
Coverage of diphones and various features
e.g. Allophonic variation
• Pin vs Spin vs limping

15. Recording Data for TTS
Tons of text
Recording script:
Few weeks of
recordings
Automatic
selection of
texts
Recording script:
• Covers all combinations of diphones
and significant features in a
language

16. an error occurred while searching for your route
because snaps weren't all so obedient anymore,
now we say apple again. and we say apple,
general electric soars today. information on general
electric
quick breads, zucchini, holiday, crock pot, cake,
so are you still keeping tabs on your old team,
that weighs more than four tons, disrupts the
herring's swim
…
An apple a day, keeps …

17. Getting started

18. Get started

19. Voicing your blog

20. AWS Blog

21. Architecture
RSS Feed Amazon Polly
Amazon
CloudWatch
Amazon S3AWS Lambda
1. Trigger
2. Check
3. Content
4. Text 5. Audio
6. Audio

22. Workshop & Demo
https://github.com/awslabs/amazon-polly-sample

23. Source
from boto3 import Session, resource
from contextlib import closing
polly = Session().client(”polly")
response = polly.synthesize_speech(
Text="Sample content",
OutputFormat="mp3", VoiceId="Joanna")
with closing(response["AudioStream"]) as stream:
bucket = resource("s3").Bucket("podcasts")
bucket.put_object(Key="output.mp3", Body=stream.read())

24. Summary
RSS Feed Amazon Polly
Amazon
CloudWatch
Amazon S3AWS Lambda
1. Trigger
2. Check
3. Content
4. Text 5. Audio
6. Audio

25. Other use cases

26. Polly is cost-effective
 Pay-as-you-go
 $4 for 1M characters
 Free Tier of 5M characters/month - first year
 You can store and reuse generated speech

27. Thank you!