Text-to-Speech (TTS)

When you provide text, Twilio synthesizes speech in real time and speaks the audio in any call. You can use TTS in TwiML and Twilio Studio.

To provide plain text that Twilio converts to synthesized speech, use the <Say> verb.

1
<Response>
2
   <Say>Hello world!</Say>
3
</Response>

1
<Response>
2
   <Say language="en-US" voice="Polly.Joanna">Hello. I am Joanna and I speak American English!</Say>
3
</Response>

<Say> offers different options for voices. Each option offers its own supported set of languages and genders. To customize your application to your needs and preferences, use the Text-to-Speech capabilities.

To start using Text-to-Speech, complete the following steps:

1. Configure your account-wide text-to-speech Settings in the Twilio Console.
2. Define TTS instructions in a TwiML document with <Say>.

To design and build applications with little or no code, use Twilio Studio. Studio uses Widgets to represent Twilio's features and functionality.

To add Text-to-Speech capabilities to your application, add the Say/Play Widget.

1. Configure your account-wide Text-to-Speech settings in the Twilio Console.
2. To include TTS in your Studio Flow, add the Say/Play Widget.

You can choose from three types of Twilio Text-to-Speech: Basic, Standard, Neural and Generative. These types differ in their support of the following parameters:

• technology used
• languages and locales supported
• quality of conversation employed
• pricing offered

These voices help you learn Text-to-Speech capabilities using <Say>. Don't use them for production applications. These voices lack enough human-like qualities for voice call conversation. Due to their limited purpose, these voices support few languages, but Twilio provides them at no additional cost.

These voices use standard technology in synthesized speech, and produce natural-sounding lifelike voices but might have limited human speech patterns and inflections. These voices speak using Amazon Polly and Google Standard voices.

These voices use enhanced technology in synthesized speech. They produce higher-quality and more natural-sounding voices than Standard voices. These voices speak using Amazon Polly Neural and Google WaveNet and Neural2 voices.

These voices are powered by the latest technology and innovation in synthesized speech to offer the most human-like, emotionally engaged and adaptive context-aware voices by "interpreting" the text-input and adjust speech accordingly (e.g. render context-dependent prosody, tone, emotion, pausing, spelling, dialectal properties, foreign word pronunciation, etc). These synthetic voices are remarkably similar to a human voice, and make them the optimal option for Conversational AI applications and Virtual Agents. These voices speak using Amazon Polly Generative and Google Chirp3-HD voices.

The following table contains all voices available for each language and locale. You can test the different voices on the Text-to-Speech page in the Twilio Console.

Bilingual voices are indentified with (*) in the Voice column. At the moment, only Amazon Polly supports this for a limited number of voices. To learn more about the bilingual voices, consult the Amazon Polly documentation.

To define the default voice and language for your account, go to the Text-to-Speech page in the Twilio Console.

• When you don't set language or voice attributes in your <Say> TwiML verb, it uses the default values.
• When you select Default in Studio, it uses the default values.

You can test different voices and messages in this section of the Console.

1
<Response>
2
  <Say>Hello. I am a man!</Say>
3
</Response>

Twilio updates the offered Text-to-Speech voices on a regular basis. To access the latest voices without needing to review and change your code, use the Language Mapping feature. Your application only needs the language and the text. Twilio automatically selects and uses the corresponding voice. You can update these at any time from the Console.

On the Text-to-Speech page in the Console, you can set a voice for every locale.

To set a voice for a locale, complete the following steps.

1. Go to the Text-to-Speech page in the Console.
2. Under the Current Language Mapping heading, click the language and locale you wish to configure. As an example, choose English (British)(en-GB). The Test & Configure Voices By Language modal displays.
3. From the dropdown menus, select the Provider and Voice you wish to use. As an example, choose Amazon Polly and Emma.
4. Click Save.
5. Repeat steps 1 to 4 for other language and voice pairing you want to use.

With these mappings set, you can specify the language without specifying the voice when using the <Say> TwiML verb in your application.

1
<Response>
2
  <Say language="en-GB">Hello. I am Emma!</Say>
3
</Response>

<Say>'s voice attribute allows you to override any default voice settings that were configured in the Console (i.e. Account-level and Language Mapping defaults).

1
<Response>
2
  <Say voice="Polly.Joanna">Hello. I am Joanna!</Say>
3
</Response>

To override a Language Mapping's defaults, use the voice attribute.

1
<Response>
2
  <Say language="en-GB" voice="Polly.Joanna">Hello. I am Joanna!</Say>
3
</Response>

<Say>'s language attribute allows you to override any default language settings that were configured in the Console (i.e. Account-level and Language Mapping defaults).

1
<Response>
2
  <Say language="de-DE">Hallo. Ich spreche Deutsch!</Say>
3
</Response>

To fine-tune synthesized speech, use SSML tags. With SSML, you can specify where pauses should be, provide pronunciations for acronyms, abbreviations, dates and times, and increase or decrease the speed of spoken text.

Twilio supports a subset of SSML tags.

SSML support may differ between Text-to-Speech providers, limited to specific voices, or both. Review the provider-specific SSML documentation and test your application. Use of SSML tags that a Text-to-Speech provider or voice doesn't support may result in error and <Say> instruction failure.

The following table lists supported SSML tags. To verify the correct use of SSML tags, consult the appropriate provider-specific documentation and test your application.

To control the volume, rate, and pitch of synthesized speech, use the SSML <prosody> tag.

1
<Response>
2
  <Say voice="Polly.Joanna">
3
    Prosody can be used to change the way words sound. The following words are
4
    <prosody volume="x-loud"> quite a bit louder than the rest of this passage.
5
    </prosody> Each morning when I wake up, <prosody rate="x-slow">I speak slowly and
6
    deliberately until I have my coffee.</prosody> I can also change the pitch of my voice
7
    using prosody. Do you like <prosody pitch="+5%"> speech with a pitch higher,</prosody>
8
    or <prosody pitch="-10%"> is a lower pitch preferable?</prosody>
9
  </Say>
10
</Response>

To indicate specific categories of text, use the SSML <say-as> tag. This ensures the correct pronunciation with synthesized speech.

Without <say-as>, the voice pronounces a phone number like a number.

This results in pronouncing 4155551212 as four billion, one hundred fifty-five million, five hundred fifty-one thousand, two hundred twelve.

To read the phone number as four one five, five five five, one two one two, use <say-as> as in the following TwiML document.

1
<Response>
2
   <Say voice="Polly.Joanna">John's phone number is, <say-as interpret-as="telephone">4155551212</say-as></Say>
3
</Response>

You can generate TwiML with SSML within the <Say> verb using one of Twilio's helper libraries for C#, Java, Node.js, PHP, Python, Ruby, or Go.

The following code sample shows helper library code that generates the following SSML and TwiML:

1
<Response>
2
  <Say voice="Polly.Joanna">
3
    Hi
4
    <break strength="x-weak" time="100ms"/>
5
    <emphasis level="moderate">Words to emphasize</emphasis>
6
    <p>Words to speak</p>
7
    aaaaaa
8
    <phoneme alphabet="x-sampa" ph="pɪˈkɑːn">Words to speak</phoneme>
9
    bbbbbbb
10
    <prosody pitch="-10%" rate="85%" volume="-6dB">Words to speak</prosody>
11
    <s>Words to speak</s>
12
    <say-as interpret-as="spell-out">Words to speak</say-as>
13
    <sub alias="alias">Words to be substituted</sub>
14
    <w>Words to speak</w>
15
  </Say>
16
</Response>

1
const VoiceResponse = require('twilio').twiml.VoiceResponse;
2

3
const response = new VoiceResponse();
4
const say = response.say({
5
    voice: 'Polly.Joanna'
6
}, 'Hi');
7
say.break({
8
    strength: 'x-weak',
9
    time: '100ms'
10
});
11
say.emphasis({
12
    level: 'moderate'
13
}, 'Words to emphasize');
14
say.p('Words to speak');
15
say.addText('aaaaaa')
16
say.phoneme({
17
    alphabet: 'x-sampa',
18
    ph: 'pɪˈkɑːn'
19
}, 'Words to speak');
20
say.addText('bbbbbbb')
21
say.prosody({
22
    pitch: '-10%',
23
    rate: '85%',
24
    volume: '-6dB'
25
}, 'Words to speak');
26
say.s('Words to speak');
27
say.sayAs({
28
    'interpret-as': 'spell-out',
29
    role: 'yyyymmdd'
30
}, 'Words to speak');
31
say.sub({
32
    alias: 'alias'
33
}, 'Words to be substituted');
34
say.w('Words to speak');
35

36
console.log(response.toString());

1
<Response>
2
  <Say voice="Polly.Joanna">
3
    Hi
4
    <break strength="x-weak" time="100ms"/>
5
    <emphasis level="moderate">Words to emphasize</emphasis>
6
    <p>Words to speak</p>
7
    aaaaaa
8
    <phoneme alphabet="x-sampa" ph="pɪˈkɑːn">Words to speak</phoneme>
9
    bbbbbbb
10
    <prosody pitch="-10%" rate="85%" volume="-6dB">Words to speak</prosody>
11
    <s>Words to speak</s>
12
    <say-as interpret-as="spell-out">Words to speak</say-as>
13
    <sub alias="alias">Words to be substituted</sub>
14
    <w>Words to speak</w>
15
  </Say>
16
</Response>

• Basic voices can process no more than 4,000 characters.
• Basic voices don't support SSML tags.

• <Say> can process no more than 3,000 characters excluding SSML.
• <Say> doesn't support Amazon-specific SSML tags. These would include <amazon:auto-breath> or <amazon:effect>.
• <Say> doesn't support lexicons.
• SSML support may vary between Polly and Polly Neural voices.
  To review any differences, consult the Amazon Polly SSML documentation.

• <Say> can process no more than 5,000 characters, including SSML, newlines and spaces.
• As Google includes SSML tags, newlines and spaces in the total character count, they count toward Google billing totals.
• <Say> doesn't support Google-specific SSML tags. This includes <par> or <seq>.
• SSML support may vary between Standard, WaveNet and Neural2 voices.
  To review any differences, consult the Google SSML documentation.